Abstract
Background Therapeutic sleep deprivation (SD) rapidly induces robust, transient antidepressant effects in a large proportion of patients suffering from a major depressive episode, but underlying biological factors remain poorly understood. Research suggests that depressed patients may have altered circadian molecular genetic ‘clocks’ and that SD functions through ‘resetting’ dysregulated genes; additional factors may be involved, warranting further investigation. Leveraging advances in microarray technology enabling the transcriptome-wide assessment of gene expression, this study aimed to examine gene expression changes accompanying SD and recovery sleep in major depression.
Methods Patients (N=78) and controls (N=15) underwent SD, with blood taken at the same time of day before, after one night of SD and after recovery sleep. A transcriptome-wide gene-by-gene approach was used, with a targeted look also taken at circadian genes. Furthermore, gene set enrichment, and longitudinal gene set analyses including the time point after recovery sleep, were conducted.
Results Circadian genes were significantly affected by SD, with patterns suggesting that molecular clocks of responders and non-responders, as well as cases and controls respond differently to chronobiologic stimuli. Notably, gene set analyses revealed a strong widespread effect of SD on pathways involved in immune function and inflammatory response, such as those involved in cytokine and especially in interleukin signalling. Longitudinal gene set analyses showed that in responders these pathways were upregulated after SD; in non-responders, little response was observed.
Conclusions Our findings emphasize the close relationship between circadian, immune and sleep systems and their link to etiology of depression at the transcriptomic level.
Introduction
Therapeutic sleep deprivation (SD) rapidly induces robust antidepressant effects in a large proportion of patients suffering from a major depressive episode(1–5). The effects of the treatment are transient as relapse is usually observed after recovery sleep. The mechanisms through which SD exerts its antidepressant effects nevertheless offer important insights into the biological factors involved in depression and antidepressant response, and have been the focus of recent research (6–10). Work in humans(11–14) as well as animals (15–17) consistently documents the effects of mistimed or insufficient sleep and sleep deprivation on circadian gene expression (such as CLOCK, ARNTL [BMAL1], PER1, PER2, PER3, etc.,) as well as on genes involved in related biological processes such as inflammatory, immune and stress response (18, 19). A prominent hypothesis about the antidepressant mechanism underlying SD is that it restores circadian rhythmicity which is often dysregulated in depression, via resetting clock gene transcription (20, 21). The well-controlled nature and rapidity of response to SD treatment (22) renders it a promising context to investigate associated biological measures such as gene expression.
While no systematic investigation of gene expression changes in depressed patients undergoing SD treatment has been conducted to date, the study of gene expression in major depressive disorder (MDD) has raised the idea that genes associated with MDD are enriched for inflammation and immune response pathways, which may be linked to the sleep disturbances observed in depression (23–25). Circadian rhythms are found in the majority of physiological processes and the immune system is no different, with alterations of these rhythms leading to disturbed immune responses (26). The immune system and circadian clock circuitry crosstalk, with immune challenges and mediators, such as cytokines, also feeding back to affect circadian rhythms(27). Cytokines, including chemokines, interferons and interleukins, are integral to sleep homeostat regulation and can modulate behavioural and physiological functions(28).
This study aimed to examine gene expression changes accompanying SD and recovery sleep in MDD using a longitudinal design, looking at changes in peripheral blood gene expression in a sample of depressed inpatients and healthy controls. Gene expression changes occurring after SD and recovery sleep, as well as associated with response and case-control status, were explored. In a hypothesis-driven approach, expression patterns of circadian genes were investigated. For a systematic search, transcriptome-wide gene-by-gene and gene set enrichment analyses were performed, while longitudinal gene set analysis (29) explored dynamics in expression trajectories at the functional gene set level.
Methods and Materials
Participants
The present sample has been described elsewhere(30). Seventy-eight inpatients (34 females; age mean ± standard deviation = 43.54 ± 14.80 years) presenting with a major depressive episode (unipolar, n = 71; bipolar I, n = 6; and bipolar II, n = 1) and on stable medication for ≥ 5 days participated in this study. Depression was diagnosed according to ICD-10 criteria. Patients were recruited from consecutive admissions to the depression unit of the Department of Psychiatry and Psychotherapy of the Central Institute of Mental Health (CIMH), Mannheim, Germany. Prescribed medication included typical and atypical antidepressants, lithium, and adjunct therapies. Fifteen healthy controls (8 females; 40.53 ± 15.90 years) with no history of psychiatric/somatic disorders were recruited online. The investigation was carried out in accordance with the Declaration of Helsinki and approved by the local ethics committee. All participants provided written informed consent following a detailed explanation of the study.
Sleep Deprivation
On Day 1, participants gave informed consent and entered the study (see Figure 1). SD was conducted from Day 2–Day 3, in small groups of 1-5 participants under staff supervision. Participants remained awake from ~0600hrs on Day 2 to 1800hrs on Day 3 (36 hours). On Day 3, participants underwent recovery sleep from 1800-0100hrs. Sleep phase advance was then carried out, shifting sleep one hour forward each day until the patient’s regular sleep pattern was reached. Controls participated alongside patients. Response was assessed by the senior clinical researcher using the Clinical Global Impression Scale for Global Improvement/Change in the afternoon on Day 3.
Data Collection
Blood samples were collected in RNA-stabilizing PAXgene tubes (Qiagen, Hilden, Germany), processed according to standard procedures, and stored at −80⍰°C until analysis. Blood was collected at the same time (between 0600–0730hrs) on Day 2 before SD (T1), on Day 3 after SD (T2) and on Day 4 (T3) (see Figure 1). The number of samples decreased over time points (Table 1) due to non-participation.
Sample preparation and analysis
RNA and Microarray Analyses
Laboratory analyses were performed using standard methods (see Supplementary Information, SI).
Gene Expression Data Pre-processing
A Custom CDF Version 20 with ENTREZ based gene definitions was used to annotate the Affymetrix GeneChip™ Human Gene 2.0 ST arrays used for gene expression profiling (31). The Raw fluorescence intensity values were normalized applying quantile normalization and RMA background correction using SAS JMP11 Genomics, version 7(SAS Institute, Cary, NC, USA). The final dataset comprised mRNA expression targeting 24,733 unique genes for each time point per participant.
Data Analysis
Analyses were conducted in R (Microsoft R Open 3.4.2). Significance was set at FDR q < 0.05.
Gene-based Analysis
Analyses were conducted using lme4 (Version 1.1-17). Linear mixed effects models with random intercepts were fit to examine gene expression differences between T1 and T2 (‘effect of time point’). Three main models were fitted: effect of time point in ‘all cases’ (M1), in ‘responders vs. non-responders’ (M2) and in ‘cases vs. controls’ (M3). In all analyses, age and sex were included as covariates.
Models were specified as follows: for each transcript, likelihood ratio tests were calculated between two models (h1 vs h0). In both models, gene expression was specified as the dependent variable, covariates were specified as fixed effects and the individual was specified as a random effect. h1 contained the comparison of interest specified as a fixed effect while h0 was a reduced model without it. That is, in M1, the ‘effect of time point’ was the only difference between h1 and h0, while in M2 and M3, the effect of interest was the interaction of the comparison group status (i.e. responder/non-responder, and case/control) and ‘effect of time point’.
Additionally, whether differences in expression levels at T1 ‘baseline’ were informative about response and disease status was examined using linear models, controlling for sex and age.
Differential Expression of Circadian Genes
To determine differential expression of circadian genes, a list of genes comprising both traditional clock genes (in order of highest signal-to-noise ratio predicting circadian rhythmicity): PER1, NR1D2,PER3, NR1D1, PER2, ARNTL, NPAS2, CLOCK, CRY2, and CRY1 as well as other top genes shown to predict circadian rhythmicity in human blood (DDIT4, CLEC4E, FKBP5, DAAM2, TPST1, IL13RA1, SMAP2, HNRNPDL, FOSL2, PER1, FLT3, CDC42EP2, TMEM88, NR1D2, RBM3) in (32) was used. Using a Monte-Carlo approach, the probability of obtaining at least the observed number of significant associations (p < 0.05 uncorrected) in a random gene set of the same size was calculated.
Gene Set Enrichment Analysis
Gene Set Enrichment Analysis (GSEA)(33) was used to determine whether differentially expressed genes offered biologically meaningful insights about SD. Ranked lists were created based on results of models M1-M3, using a signed log10 transformed p-value with sign denoting direction of change, as described elsewhere(34). To allow a concise interpretation of the potentially widespread effects of SD interventions, the Hallmark gene set collection(35) (MSigDB Version 6.2), comprising 50 gene sets representing specific well-defined biological states/processes displaying coherent expression, was used(35). The heme metabolism gene set was excluded due to the globin interference artefact (communication with ThermoFisher).
Longitudinal Gene Set Trajectory
Time-course Gene Set Analysis (TcGSA)(29) (Version 0.12.1) was employed to examine gene expression dynamics over all three time points. TcGSA, employing mixed models, detects gene sets in which expression changes over time, taking between-gene and individual variability into account, with higher sensitivity and better interpretability than univariate individual gene analysis (for details, see (29)). As above, the Hallmark gene set collection was used. Significance was set at FDR q < 0.05. TcGSA was employed separately for all cases, controls, responders and non-responders. Data at T3 from one patient not following recovery sleep protocol was excluded.
Results
Gene-based Analysis
In all cases [M1], 4,071 (2,083 up, 1,988 down) genes were significantly differentially expressed after SD. Significant differential changes in gene expression after SD between responders and non-responders [M2] were observed in 360 genes [150 up, 210 down] and in cases vs. controls [M3] in 495 genes [248 up, 247 down].
Table 2 shows top differentially expressed genes for these models. Tables S1.1-S1.3 show the number of genes differentially expressed, upregulated, and downregulated for all models, and detailed lists of differentially expressed genes.
At baseline, no genes were significantly differentially expressed between cases/controls and responders/non-responders (Table S1.4).
Differential Expression of Circadian Genes
We observed significant differential expression of circadian genes in models M1, M2 and M3. (Table 3, for more details, see Tables S2.1-3). Baseline differences in circadian genes between responders/non-responders, and cases/controls did not reach FDR q < 0.05, but achieved nominal significance (see Table S2.4).
Monte Carlo simulations confirmed that significantly more circadian genes were differentially expressed than random gene sets of the same size (see SI).
Gene Set Enrichment Analysis
GSEA notably found enrichment in immune response related pathways (see Tables S3.1-3).
For M1, M2 and M3, 12, 23 and 11 gene sets were significantly positively enriched, respectively (FDR q < 0.05). The TNFα signalling via NFKβ gene set had the strongest positive enrichment in all models, while Inflammatory Response was also consistently among the significantly positively enriched gene sets.
Given the enrichment observed in TNFα and immune pathways, selected genes prominent in immune processes(25) were further examined (Table S4).
Time course Gene Set Analysis
TcGSA results mirrored and extended gene-based analysis results
In responders, 48 gene sets varied significantly (the model for one gene set, Complement, did not converge) (see Figure 2a). Descriptively, in comparison to T1, responders showed a spectrum of differential gene expression at T2 and maintained until T3, with strong upregulation observed (Interferon Gamma Response, Inflammatory Response, TNFα Signalling via NFKβ, IL6-JAK-STAT3-Signaling, Interferon Alpha Response and Unfolded Protein Response).
In non-responders, 39 gene sets varied significantly (see Figure 2b), the majority of gene sets were downregulated at T2, with pronounced upregulation of the abovementioned gene sets only observed at T3.
In cases, expression was observed to vary significantly in 49 gene sets (see Figure 2c). Descriptively, upregulation was observed in immune system related gene sets in two main patterns; (1) strong upregulation at T2, sustained until T3 (i.e. Inflammatory Response, IL6-Jak-Stat3-Signalling and TNFα Signalling via NFKβ) and (2) light upregulation at T2 followed by stronger upregulation at T3 (e.g., Interferon Gamma Response, Interferon Alpha Response, Apoptosis).
In controls, 27 gene sets varied significantly (see Figure 2d). Descriptively, at T2, immune/inflammation related gene sets were upregulated, with stronger upregulation observed at T3 (e.g. Interferon Gamma Response, Inflammatory Response, IL2-Stat5-Signalling, TNFα Signalling via NFKβ). The majority of other gene sets were downregulated at T2, followed by upregulation at T3.
Tables S5.1-4 list significant TcGSA gene sets.
Discussion
Here, we report the first longitudinal transcriptome-wide study of SD treatment in patients suffering from a major depressive episode. Widespread differential gene expression was observed after SD; circadian genes were differentially expressed, and enrichment in pathways related to immune function, inflammatory response and sleep regulation was observed.
It has been hypothesized that SD exerts its antidepressant effect by resetting disturbed clock gene functioning thought to be a feature of depression(9). Consequently, we examined changes in expression of circadian genes. All individuals, i.e. cases and controls, showed significant differential expression of circadian genes after SD. In responders, many significant gene expression changes were observed while in non-responders only a few nominally significant changes were found. The following observation may be of interest: three circadian genes, PER1, CLEC4E and SMAP2 (FDR q = 0.083 approaching significance, 0.048, and 0.039, respectively) showed strong differential gene expression, i.e. increased expression, in responders versus non-responders after SD. While the roles of CLEC4E and SMAP2 are yet unexplored in this context, similar PER1-related findings have been previously reported; consistent with the present findings, PER1 expression is observed to be increased in SD responders and decreased in non-responders(9). Animal studies have shown that sleep deprivation or prolonged wakefulness enhanced Per1 expression in several brain regions (36, 37), and that quetiapine increased Per1 expression in the amygdala(38). The present findings cannot yet demonstrate that clock gene dysregulation/normalization is at the core of the SD mechanism. However, they suggest that (genes of) the molecular clocks of responders/non-responders, as well as cases/controls respond differently to chronobiologic stimuli, which may be associated with treatment outcomes.
Among the top 10 genes most significantly differentially expressed after SD in cases [M1] were: i) KLF6, observed to be upregulated in individuals after experimental restriction in a genome-wide association study of sleep duration(39); (ii) SIPA1L1, expression of which was found to be increased in a study examining changes in military personnel at baseline and after improvement of sleep(40); (iii) NHSL2, the function of which is unknown, but which is located in a genomic region found to be differentially hydroxymethylated in a study of sleep deprivation in mice(41); (iv) TREM1, which has immune function(42); and (v) TSPAN2, of which a study looking for gene expression changes under fluoxetine in rats found hippocampal upregulation(43). Among the top 10 genes observed in responders vs non-responders [M2] is BECN1 (Beclin1) – which (together with FKBP5, see below) is shown to be involved in priming autophagic pathways to set the stage for antidepressant action(44, 45). Circadian rhythms of autophagic proteins, including beclin1 have been linked to sleep disturbances(46). Also in the top 10 in M2 were DNER, and GSR which have functions related to immune response(47, 48). Between cases and controls [M3], top genes included EZH2, which is reported to have a close relationship to IL-6(49). ZBTB16 is implicated in human sleep duration(39) and sleep deprivation in animal studies(50), and CISH is shown to be involved in immune related processes(51, 52). Among other differentially expressed genes are ones evidently associated with antidepressive intervention, such as FKBP5 (53) where we observed significant upregulation in cases vs. controls after SD (see Tables S1.1, S2.1). FKBP5 plays an important regulatory role in stress response (53–56); circadian rhythm abnormalities have been linked to the stress response system, and circadian clock genes can both regulate and be regulated by rhythms of glucocorticoid release(57).
Pathway analyses showed a global effect of SD on gene expression; immunological, inflammatory response and sleep regulation involved pathways were most strongly affected. These findings are of interest in light of previous gene expression studies linking immune function to MDD(58, 59); SD may affect pathways involved in the MDD etiology. In cases and especially responders, in the Inflammatory Response (genes related to cytokines, growth factors, cell differentiation markers and transcription factors)(35), IL6-JAK-STAT3-Signalling (aberrantly hyperactivated in patients with cancer and chronic inflammatory conditions) (60), and TNFα Signalling via NFKβ (cell proliferation, differentiation, apoptosis, neuroinflammation mediated cell death) gene sets, strong upregulation was observed at T2 and maintained until T3. Prior findings have shown associations between the immune system and depression, suggesting that causal pathways exist from immune dysregulation and inflammation to MDD(23, 24, 61–64). This raises the question of how SD and the immune system interact and whether SD counteracts and/or enhances depression-related immunological processes.
Interestingly, and in contrast to responders, mainly weaker downregulation was observed at T2 in non-responders, with upregulation of immune related gene sets only observed at T3. This differential function of immune related processes (i.e. blunted system responsiveness) might be associated with non-response to treatment. This preliminary finding must be further explored in larger sample sizes.
Of note, downregulation of MYC targets pathways after SD was consistently observed in both responders and non-responders but not controls. MYC, well known as an oncogene(65), acts as a mediator and coordinator of cell behaviour which also inversely modulates the impact of the cell cycle and circadian clock on gene expression(66). It has been shown that dysregulation of MYC disrupts the molecular clock by inducing dampened expression and oscillation of CLOCK-BMAL1 master circadian transcription factor(67). Removal of clock repression by MYC may play a role in the interaction of SD with depression and further investigation of these potential mechanisms is warranted.
Sleep and immunity are connected by anatomical and physiological bases(68). The role of cytokines in the brain is complex and remains to be fully understood; while a full discussion of their roles is beyond the scope of this work, Table S4 shows cytokines and substances which have been implicated in both immune and sleep regulatory processes(25). The present findings, observing upregulation after SD, support reports linking levels of pro-inflammatory cytokines to depression and suggesting involvement in disease pathogenesis(25, 69, 70).
One important clue to SD response may lie in the observed TNFα expression patterns. TNFα is a pro-inflammatory cytokine controlling expression of inflammatory genetic networks; in addition to many immune-related functions in the brain, it influences whole organism function, including sleep regulation (25, 62). After SD, cases had significantly decreased expression of TNF compared to controls (FDR q = 0.01); upregulation was observed in controls and downregulation in cases. Upregulated TNF in controls is consistent with reports of sleep deprivation-induced increases in TNFα levels in healthy people (71). The decrease observed in cases, and especially responders, may inform the mechanism of SD response- increased TNFα concentrations are reported to be a marker of depression and TNFα administration is reported to induce depressive symptoms(72).
Sleep-wake cycles are accompanied by changes in circulating immune cell numbers and disturbing the circadian cycle affects immune response (73, 74). Acute SD reportedly affects diurnal rhythmicity of cells, mirroring the immediate immune stress response(75, 76). Consistent with a stress-like response, upregulation of expression of IL6 and IL1B, thought to be involved in depression(77) and regulation of circadian rhythm/sleep homeostasis (12, 70, 78), was observed after SD. Given the tight coupling between circadian and immune systems, it may be that like the ‘resetting’ of circadian genes, the altered rhythmicity of the stress immune response system in depression is transiently normalized, leading to antidepressant effects; chronotherapeutic approaches that extend SD effects may be prolonging this normalization(20).
There are several limitations to our study. The sample sizes of non-responders and controls were limited, with less power potentially contributing to the lower number of significant genes. While the statistical approach (linear mixed models) used in the single-gene analysis and underlying TcGSA is robust against unbalancedness and missing data, caution should be exercised when considering the comparative results. Next, although blood was sampled once per day on three consecutive days, a significant but manageable load for patients, the amount of data is sparse for statistical estimation. To better leverage longitudinal data and methods, a denser sampling scheme (e.g. every few hours) will be required to attain a more refined understanding of underlying SD mechanisms. The sample studied was a naturalistic sample recruited from consecutive admissions and treated following standardized clinical guidelines. Nevertheless, strong effects were observed, perhaps in part attributable to the consistent methodology for applying SD treatment in a relatively large cohort for SD. Considering the fact that depression is a heterogeneous phenotype accompanied by a heterogeneous immunophenotype(62), means that even larger sample sizes will be needed to substantiate the present findings. Finally, we investigated peripheral tissue transcriptome-wide expression changes; although an easily obtained, valuable proxy, the correlation of expression in the blood with expression in the brain, where depression is thought to act, is imperfect and requires further study(79). On the other hand, it is precisely these cells that best represent the current status of the immune system and the inflammatory response, the gene expression of which appears at the centre of SD. In addition, the crosstalk of the organs and biorhythm tuning is also mediated via the blood system.
Our findings affirm and emphasize the close relationship between circadian, immune and sleep systems in depression at the transcriptomic level, but the directionality of cause-effect remains unclear. Circadian, immune and sleep dysregulation may precede, accompany or come as a result of depression; they now represent targets for treatment which have the ability to influence clinical outcomes. Closer investigation of these systems with larger sample sizes and denser, longer-term sampling schemes will be key to disentangling and understanding the multi-level interactions occurring.
Disclosures
AM-L has received consultant fees from Boehringer Ingelheim, Brainsway, Elsevier, Lundbeck Int. Neuroscience Foundation and Science Advances. All other authors report no biomedical financial interests or potential conflicts of interest.
Acknowledgements
This work has been supported by the Deutsche Forschungsgemeinschaft (DFG), SCHW 1768/1-1. AM-L acknowledges support from the German Federal Ministry of Education and Research (BMBF, grants 01ZX1314G, 01GS08147, 01EF1803A, 255156212 CRC 1158 subproject B09) and the Ministry of Science, Research and the Arts of the State of Baden-Wuerttemberg, Germany (MWK, grants 42-04HV.MED(16)/16/1, 42-04HV.MED(16)/27/1). MR was supported by the BMBF through grants within the e:Med research program 01ZX1314G. This manuscript was uploaded to the preprint server bioRχiv.