Changes in brain gene expression after long-term sleep deprivation


Address correspondence and reprint requests to Chiara Cirelli, Department of Psychiatry, University of Wisconsin, Madison, 6001 Research Park Boulevard, Madison, WI 53719, USA.


Long-term sleep deprivation in rats produces dramatic physiological changes including increase in energy expenditure, decrease in body weight, and death after 2–3 weeks. Despite several studies, the sleep deprivation syndrome remains largely unexplained. Here, to elucidate how prolonged sleep loss affects brain cells we used microarrays and screened the expression of > 26 000 transcripts in the cerebral cortex. Rats were sleep deprived using the disk-over-water method for 1 week. Seventy-five transcripts showed increased expression in these animals relative to controls that had been spontaneously awake or sleep deprived for a few hours. Most of them were induced as a result of chronic sleep loss and not non-specific effects of the disk stimulation. They include transcripts coding for several immunoglobulins, stress response proteins (macrophage inhibitor factor-related protein 14, heat-shock protein 27, α-B-crystallin), minoxidil sulfotransferase, globins and cortistatin. Twenty-eight transcripts decreased their expression in long-term sleep-deprived rats. Sixteen of them were specifically decreased as a result of chronic sleep loss, including those coding for type I procollagen and dihydrolipoamide acetyltransferase. We also compared sleeping rats to short-term and long-term sleep-deprived rats, and found that acute and chronic sleep loss led to some differences at the molecular level. Several plasticity-related genes were strongly induced after acute sleep deprivation only, and several glial genes were down-regulated in both sleep deprivation conditions, but to a different extent. These findings suggest that sustained sleep loss may trigger a generalized inflammatory and stress response in the brain.

Abbreviations used

acetylcholine receptor


immunoglobulin heavy chain-binding protein


brain-derived neurotrophic factor


α-B crystallin


D site albumin promoter-binding protein






expressed sequence tag


gc-Robust Multi-Array


heat-shock protein


long-term sleep deprived


myelin-associated glycoprotein


macrophage inhibitor factor-related protein


nerve growth factor


non-rapid eye movement


quantitative PCR


rapid eye movement


Rnase protection assay




suprachiasmatic nucleus


short-term sleep deprived


spontaneously awake


yoked control

Sleep and wakefulness are associated with widespread changes in brain gene expression in both vertebrates (Cirelli et al. 2004; Terao et al. 2005) and invertebrates (Cirelli et al. 2005a). In the rat cerebral cortex, ∼ 5% of the transcripts are differentially expressed after 8 h of spontaneous sleep compared with 8 h of either spontaneous wakefulness or sleep deprivation (Cirelli et al. 2004). Sleep-related transcripts code for glial proteins, proteins involved in protein synthesis, cholesterol synthesis, membrane trafficking, synaptic down-regulation and memory consolidation. Transcripts expressed at higher levels during spontaneous wakefulness or short-term sleep deprivation, on the other hand, include those coding for several mitochondrial proteins, chaperones and heat-shock proteins, and proteins involved in synaptic potentiation and glutamatergic transmission. These findings suggest that continuous wakefulness not only increases brain energy demand, but also represents a cellular stress for neurons and/or glial cells. Indeed, a recent study found that, in the mouse cerebral cortex, as little as 6 h of forced wakefulness triggers the unfolded protein response, which includes the slowing down of protein synthesis and the induction of immunoglobulin heavy chain-binding protein (BiP) (Naidoo et al. 2005).

Long-term sleep deprivation in rats produces a series of dramatic physiological changes that invariably culminate in death after 2–3 weeks (Rechtschaffen and Bergmann 2002). Within the first 1–2 days, sleep-deprived rats show an increase in food intake, energy expenditure and heart rate, followed 1–3 weeks later by a decrease in body weight and in body and brain temperature. The sleep deprivation syndrome and its lethal consequences have also been observed after selective rapid eye movement (REM) sleep deprivation, although the pathology associated with the loss of sleep takes longer to appear, the survival time is longer, and body and brain temperature are not significantly decreased (Rechtschaffen and Bergmann 2002). Despite extensive studies, the long-term sleep deprivation syndrome has not been fully explained (Rechtschaffen 1998). No major organ pathology has been documented, and systemic infections, although a frequent and early event in sleep-deprived animals (Everson and Toth 2000), have been prevented without reversing the time course of the syndrome (Bergmann et al. 1996). In the brain, signs of brain cell death have either not been found (Cirelli et al. 1999) (Hipolide et al. 2002) or found only in the supraoptic nucleus of the hypothalamus (Eiland et al. 2002). Moreover, markers of oxidative stress are either absent (e.g. in the cerebral cortex; Gopalakrishnan et al. 2004) or restricted to a few brain areas (e.g. the hippocampus; D'Almeida et al. 1998; Ramanathan et al. 2002). Finally, even when markers of oxidative stress are induced, there is no sign of oxidative damage, including protein oxidation, lipid peroxidation and nucleic acid oxidation (D'Almeida et al. 1997, 1998; Gopalakrishnan et al. 2004). By contrast, episodic hypoxia without sleep loss is sufficient in rodents to cause cortico-hippocampal apoptosis and cognitive impairment, at least partially by inducing oxidative stress (Xu et al. 2004). This suggests that any potentially irreversible damage in the brain of patients with obstructive sleep apnea is more likely the result of abnormalities in blood gas composition than to sleep loss per se.

The goal of this study was to elucidate the molecular consequences of prolonged sleep loss on brain cells. With this aim, we used cDNA microarrays to screen the expression of > 26 000 transcripts in the cerebral cortex of rats sleep deprived for 1 week using the disk-over-water (DOW) method. The DOW is the most effective and best controlled system for enforcing long-term sleep deprivation in animals (Rechtschaffen and Bergmann 2002). The cerebral cortex was chosen because it is responsible for the cognitive defects observed after sleep deprivation (Horne 1988;Van Dongen et al. 2003), and it is at the center of most hypotheses concerning the functions of sleep (Moruzzi 1972; Horne 1988; Krueger et al. 1995; Maquet 1995; Steriade and Timofeev 2003; Tononi and Cirelli 2006).

Materials and methods

Experimental groups and polygraphic recordings

Under pentobarbital anesthesia (75 mg/kg, intraperitoneal), adult male Wistar Kyoto rats (300–450 g, n = 52) were implanted with screw electrodes in the skull to record the electroencephalogram (EEG), and with silver electrodes in the nuchal and temporal muscles to record the electromyogram. After surgery, rats were housed individually in sound-proof recording cages where lighting and temperature were kept constant (light : dark 12 : 12 h, light on at 10.00 hours; 25°C ± 1°C; food and drink ad libitum). Each day from 10.00 to 10.30 hours all rats were handled and exposed to novel objects to become familiar with the procedure for short-term sleep deprivation (see below). One week after surgery, the rats were connected by means of a flexible cable and a commutator (Airflyte, Bayonne, NJ, USA) to an electroencephalograph (model 15LT; Grass, West Warwick, RI, USA) and recorded continuously for as many days (7–30 days) as required to satisfy the criteria for each of the five experimental groups. EEG signals were scored visually for 4-s epochs (SleepSignTM; Kissei Comtec America, Inc., Irvine, CA, USA). Long-term sleep-deprived (l-SD) rats were kept awake using the DOW method for 7 days, killed between 10.00 and 11.00 hours, and compared with their yoked controls (YCs; see below). One week of sleep deprivation was selected because it is sufficient to induce all the physiological markers of the sleep deprivation syndrome (Rechtschaffen and Bergmann 2002). The remaining three groups included sleeping (S), short-term sleep-deprived (s-SD) and spontaneously awake (W) rats. The animals in the last three groups were the same as used in a previous study (Cirelli et al. 2004). S rats were killed during light hours (between 17.00 and 18.00 hours), at the end of a long period of sleep (> 45 min, interrupted by periods of wakefulness of < 2 min), and after spending at least 75% of the previous 7–8 h asleep. Rats in the s-SD group were kept awake for the first 8 h of their sleeping period by introducing novel objects into their recording cages. They were killed at the same circadian time as the S group to assess the effects of behavioral state independently of circadian factors. W rats were killed during the dark period (between 05.00 and 06.00 hours), at the end of a long period of wakefulness (> 1.5 h, interrupted by periods of sleep of < 5 min) and after spending at least 70% of the previous 7–8 h awake. Animal protocols followed the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the University of Wisconsin.

Sleep deprivation procedure

Short-term sleep deprivation was enforced by exposing the rats to novel objects. Every new object was introduced into the cages just following the first signs of slow waves in the frontal EEG signal. Long-term sleep deprivation was performed by the DOW method (Rechtschaffen and Bergmann 2002). Briefly, a rat that was to be sleep deprived and its YC were housed in rectangular Plexiglas cages. A single horizontal disk 46 cm in diameter, which could be rotated in a randomly chosen direction, formed a floor extending 17 cm into each cage. Under the disk and extending to the cage walls was a rectangular tray filled with tap water to a depth of approximately 2 cm. When sleep onset was detected in the l-SD rat, the disk was rotated slowly by a computerized monitoring system, forcing both rats to walk in a direction opposite to disk rotation to avoid the water. When the l-SD rat was spontaneously awake, the disk was stationary and the YC rat was able to sleep. During baseline, the disk was rotated once per hour for 6 s to habituate the rats to rotation. The baseline period continued until sleep, food intake, body weight and temperature had stabilized (usually 3–7 days) in both rats. Cage air temperature was thermostatically maintained at 25°C ± 1°C.

Microarray: labeling, hybridization and data analysis

Rats were deeply anesthetized with isoflurane (within 2 min) and decapitated. The head was cooled in liquid nitrogen and the whole brain was removed. Cerebral cortex, hippocampus and cerebellum were dissected, and the rest of the brain was left intact. Samples were immediately frozen on dry ice and stored at − 80°C. Total RNA was isolated from the entire right cerebral cortex of each animal by using Trizol (Gibco-BRL, Gaithersburg, MD, USA) according to the manufacturer's instructions. Final RNA concentrations were determined spectrophotometrically. Equal mass amounts of total RNA from the right cerebral cortex of rats in the same experimental group were pooled (n = 6 rats for the S, s-SD, W groups; n = 7 for l-SD and YC). The pools (20 µg/pool) were used to hybridize three to five (S, s-SD, and W groups) or five to eight (l-SD and YC groups) independent sets of Affymetrix Rat Genome Arrays (U34A,B,C) containing more than 7000 annotated sequences and 18 000 expressed sequence tags (ESTs). Microarray labeling, hybridization and expression analyses were performed according to the Affymetrix GeneChip Expression Analysis Technical Manual (Affymetrix Inc., Santa Clara, CA, USA) and essentially as previously reported (Cirelli et al. 2004). Briefly, an equal mass amount of total cortical RNA from each pool was converted into first-strand cDNA using Superscript II RNAse H- reverse transcriptase (Invitrogen Life Technologies, Carlsbad, CA, USA) and the second strand was synthesized. cDNA was then converted to biotinylated cRNA using the ENZO BioArray High Yield In Vitro Transcription kit (Enzo Life Sciences, Farmingdale, NY, USA) according to the manufacturer's instructions. The cRNA was fragmented at 0.5 µg/mL final concentration in 1 × fragmentation buffer (40 mm Tris-acetate, pH 8.1, 100 mm potassium acetate, 30 mm magnesium acetate). The size range of cRNA before (0.5 kb and longer) and after (35–200 base fragments) fragmentation was checked by denaturing agarose electrophoresis. The quality of cDNA and cRNA syntheses was determined by the 3′/5′ ratio of rat housekeeping genes within the array (genes coding for ubiquitin, rat glyceraldehyde-3-phosphate dehydrogenase, β-actin and hexokinase).

Microarray labeling and hybridization for all five experimental groups were done simultaneously, and the results of the expression analyses for the S, s-SD and W animals have been previously published (Cirelli et al. 2004). Here, to identify transcripts specifically regulated after long-term sleep deprivation, expression measures for each probe set for all five experimental groups were calculated using the robust multiarray average gcRMA measure (Irizarry et al. 2003) and input into GeneSpring 7.2 software (Silicon Genetics, Redwood City, CA, USA) for further analysis. The original data set, which contains 26 261 probe pairs, was scrubbed to retain only those (17 169) whose absolute, non-normalized values were higher than 15 (value range 15–42 100) in at least two arrays. Statistical significance was determined by condition-to-condition comparisons requiring a significance level of 95% or greater (Welch t-test). Transcripts specifically up- or down-regulated after long-term sleep deprivation were identified using a conjunction analysis, and were defined as those whose expression increased or decreased respectively in l-SD rats relative to s-SD rats, as well as in l-SD rats relative to W rats. They are listed in Tables 2 and 3, and in Supplementary. Because in our hands real-time quantitative PCR (qPCR) has a global absolute error of ∼ 20%, we only report in the tables transcripts for which the fold change in each comparison was at least 21%.

Table 2.   Transcripts up-regulated in long-term sleep deprivation relative to short-term sleep deprivation and spontaneous wakefulness
GenBankGene symbolGene description% change
% change
Up in l-SD vs. YC
Up in YC
Up in waking
Up in sleep
  1. Rat cortical genes whose mRNA levels are increased (> 1.2-fold change, Welch t-test, p < 0.05) in long-term sleep deprivation (l-SD) relative to 8 h of sleep deprivation (s-SD) and 8 h of spontaneous wakefulness (W). % change (array), percentage change in l-SD relative to s-SD, and l-SD relative to W, respectively, as indicated by array analysis. Up in l-SD vs. YC, percentage change in l-SD relative to YC, as indicated by array analysis. Up in YC, percentage change in YC relative to s-SD, and YC relative to W, respectively, as indicated by array analysis. Up in waking (s-SD and W > S) and up in sleep (S > s-SD and W), percentage change as indicated by array analysis. % change (PCR) indicates the transcripts confirmed using qPCR (l-SD vs. s-SD, mean ± SEM % increase).

DNA binding/regulation of transcription
 J03179DbpD site albumin promoter-binding protein40, 57  40, 57 50, 71
 AI233425Atf5activating transcription factor 544, 62     
 D82074Neurod1neurogenic differentiation factor21, 71 21   
 Z38067c-mycDNA for c-myc, exon 221, 21     
 U17837RIZzinc finger protein RIZ22, 22  22, 22  
 U17837-atRIZzinc finger protein RIZ21, 31  21, 31  
 M18527IgKIg germline κ-chain C-region gene128, 433 60   
 M18532IgKIg germline κ-chain C-region gene220, 300 60   
 M28671IgG-2bESTs, Highly similar to GCB_rat Ig γ-2B chain100, 900 100   
 U39609anti-NGF30anti-NGF30 antibody light chain283, 4751800 ± 18291   
 M28670IgG1rearranged γ-1 gene, last 4 exons257, 525 150   
 U75411 anti-idiotype immunoglobulin M light chain gene114, 200 50   
 M18529IgKIg germline κ-chain C-region gene180, 600 40   
 M18531IgKIg germline κ-chain C-region gene114, 650 50   
 M12981IgKIg germline κ-chain C-region gene183, 466 54   
 M18530IgKIg germline κ-chain C-region gene114, 200 50   
 M18526IgKIg germline κ-chain C-region gene166, 700 60   
 M18528IgKIg germline κ-chain C-region gene425, 2000 110   
 M12822IgKIg germline κ-chain C-region gene300, 1900 100   
 L22654anti-AchRanti-AchR antibody gene525, 24001249 ± 20110200, 1100  
 AA850138 Ig active lambda2-like chain71, 100 20   
Stress response
 L18948MRP14MRP14, S100 calcium-binding protein A9460, 250960 ± 32100   
 AI176658Hspb1Heat shock 27 kDa protein21, 21   21, 21 
 AA818604Hspa1aheat shock 70 kDa protein 1 A60, 60  50, 5525, 25 
 X60351Cryabalpha B-crystallin25, 25   28, 28 
 AI102739GIIg15bglobal ischemia-induced protein GIIG15B38, 38  38, 38  
 J03752Mgst1microsomal glutathione S-transferase 133, 33  33, 33  
 AI172411Gpx3plasma glutathione peroxidase precursor22, 57     
 M11794 metallothionein-2 and metallothionein-1 genes21, 25     
 U39206CYP4F4cytochrome P450 4F430, 30     
 X56325 2-α-1 globin gene55, 55 30 21, 21 
 AI236360Hba1hemoglobin, α155, 55 40 50, 50 
 AI179971Hba1hemoglobin, α121, 33     
 M94919Hbbβ-globin gene50, 3542 ± 535 29, 43 
 L19998Sult1a1minoxidil sulfotransferase, sulfotransferase family 1A28, 40 40 83, 67 
 D42148Gas6growth potentiating factor27, 27     
 M63656Aldocaldolase C, fructose-biphosphate22, 38  22, 38  
 M12112 angiotensinogen33, 50  22, 38  
 AJ007627Kcnh3potassium voltage-gated channel21, 30  21, 21  
 X97121Ntsr2neurotensin receptor 221, 28     
 AF082160Grifingalectin-related interfiber protein81, 400 33   
 M15880NpyNeuropeptide Y33, 50  33, 50  
 U51919Cortcortistatin60, 33     
 L06040Alox12arachidonate 12-lipoxygenase63, 63     
 M95591Fdft1squalene synthetase24, 21  21, 21 30, 22
 U10357Pdk2pyruvate dehydrogenase kinase 226, 50     
 J02679Nqo1NAD(P)H dehydrogenase, quinone 122, 38     
 AF034582VAP1vesicle-associated protein21, 40     
 AI044247glbdiacetyl/l-xylulose reductase31, 31     
 AI103584Klf15Kruppel-like factor 1550, 33  63, 44  
 AI230514Ilf3interleukin enhancer binding factor 321, 21     
 AI070653Rpl21ribosomal protein L2130, 63     
 AI639275Tcp11ESTs21, 25     
 AI009141 ESTs50, 21     
 AI237378 ESTs28, 35  28, 35  
 AA799471 ESTs75, 100     
 AA875577 ESTs71, 50  57, 38 57, 38
 AI011556  21, 25     
 AA891914Acy1ESTs21, 25     
 AI060207Sf3b1ESTs50, 33  50, 33  
 AA851497  44, 44     
 AA955531 ESTs41, 25     
 AA848826 ESTs26, 26     
 AA963698 ESTs28, 35     
 AA943273 ESTs22, 38     
 AI031043 ESTs40, 31    33, 25
 AI101739VipESTs30, 44     
 AI230970  40, 75     
 AI178093Fxyd3ESTs21, 21     
 AI171607 ESTs30, 62     
 AI171916 ESTs38, 38  25, 25 25, 25
 AI111674 ESTs44, 30 30   
 AI112428Ctbp2ESTs33, 50     
 AI072292 ESTs30, 30     
 AA858528 ESTs21, 50 33   
Table 3.   Transcripts down-regulated in long-term sleep deprivation relative to short-term sleep deprivation and spontaneous wakefulness
GenBankGene symbolGene description% change
% change
Down in l-SD vs.YC
Down in YC
Up in waking
Up in sleep
  1. Rat cortical genes whose mRNA levels are decreased (> 1.2-fold change, Welch t-test, p < 0.05) in long-term sleep deprivation (l-SD) relative to 8 h of sleep deprivation (s-SD) and 8 h of spontaneous wakefulness (W). % change (array), percentage change in l-SD relative to s-SD, and l-SD relative to W, respectively, as indicated by array analysis. Down in l-SD vs. YC, percentage change in l-SD relative to YC, as indicated by array analysis. Down in YC, percentage change in YC relative to s-SD, and YC relative to W, respectively, as indicated by array analysis. Up in waking (s-SD and W > S) and up in sleep (S > s-SD and W), percentage change as indicated by array analysis. % change (PCR) indicates the transcripts confirmed using qPCR (l-SD vs. s-SD, mean ± SEM % decrease).

Immediate early genes
 AB003726Homer1ahomer, neuronal immediate early gene, 155, 4166 ± 5 55, 41340, 240 
 AI172476TiegTGFB inducible early growth response37, 21  50, 21128, 42 
 U78102Egr2early growth response 221, 29  21, 29500, 600 
Cytoskeleton related
 U75405Col1a1α1 type I collagen56, 50  67, 62  
 M27207Col1a1α1 type I collagen56, 4367 ± 1 62, 50  
 AI101443Col1a2procollagen, type I, α245, 55  56, 67  
 AA819207Col1a2procollagen, type I, α259, 42 23   
 X17012Igf2insulin-like growth factor II31, 21  39, 27  
 AF062740PDP1pyruvate dehydrogenase phosphatase isoenzyme 121, 21  21, 2163, 25 
 AF062740-PDP1pyruvate dehydrogenase phosphatase isoenzyme 130, 21  25, 2175, 38 
 AI172135Dlatdihydrolipoamide acetyltransferase27, 21     
 AI145931Uae1UDP-N-acetylglucosamine-2-epimerase/21, 21  21, 2150, 44 
 AI234822Rasd1DEXRAS1 (Dexras1)56, 27  45, 21157, 57 
 AI235890RT1.S3BM1k MHC class Ib antigen40, 25  33, 21  
 U60063Aldh1a2aldehyde dehydrogenase family 1, subfamily A221, 21  25, 29  
 A09811 BRL-3A binding protein33, 27  33, 27  
 AI169327Timp1tissue inhibitor of metalloproteinase 131, 21  31, 21  
 AA957564Cyp26b1cytochrome P450, family 26, subfamily b, polypeptide 127, 33     
 AA926109 ESTs66, 50  66, 50200, 100 
 AA891834 ESTs27, 32  23, 29  
 AI012434 ESTs25, 31  25, 31  
 AA893607 ESTs31, 21  39, 27  
 AI178527 ESTs27, 21 21   
 AI112149Edg2ESTs21, 43 27   
 AI237525 ESTs21, 21     
 AI235948NidESTs21, 29  25, 36  
 AI179677 ESTs21, 31     
 AI170394Mpeg1ESTs27, 33     

Real-time qPCR and Rnase protection assay (RPA)

Microarray data were confirmed using qPCR and RPA. Real-time qPCR (Sequence Detection System 5700; Perkin Elmer, Wellesley, MA, USA) was performed as previously described (Cirelli et al. 2004). RT reactions were carried out in parallel on Dnase I-digested, pooled total RNA from four new rats for each group not used for the array study (independent pool). PCR was done in quintuplicate for each sample condition assayed; mean ± SEM values are shown in the tables) and relative quantities determined based on the equation of the line of best fit derived from the standard curve (R2 ≥ 0.985). qPCR was run on a total of 16 transcripts, including three ‘controls’ that did not change according to the microarrays (coding for EF2, EIF2a, ERP72): they were confirmed as ‘non-changing’ by qPCR. Four transcripts were identified by the microarray analysis as up-regulated in l-SD rats [coding for anti-nerve growth factor (NGF)30, anti-acetylcholine receptor (AchR), macrophage inhibitor factor-related protein (MRP)14, β-globin; Table 2] and nine were identified as down-regulated in l-SD rats [coding for Homer1, collagen α1, cytochrome p450 (Cyp)26b1, Plp, myelin-associated glycoprotein (MAG), Mog, Gria, Grin2b, Cd9; Table 3, Table S4]. They were all confirmed except the transcript for MAG.

For RPAs, antisense RNA probes complementary to the coding region of the gene for brain-derived neurotrophic factor (BDNF) (Maisonpierre et al. 1991), Arc (Lyford et al. 1995), BiP (Haas and Meo 1988) and VGF (Ferri et al. 1992) were synthesized by run-off transcription from a linearized DNA template using the MAXIscript in vitro transcription kit (Ambion, Austin, TX, USA) and [α-32P]UTP (New England Nuclear-Du Pont, Natick, MA, USA). RPA was performed using the RPAIIITM kit (Ambion). Dnase I-digested total RNA from the right cerebral cortex was hybridized with an excess of [α-32P]UTP-labeled riboprobes. To normalize the amount of sample RNA, a β-actin riboprobe was used to measure β-actin mRNA. Transcripts for qPCR and RPA confirmation were selected based on multiple criteria including (i) the fold change after long-term sleep loss (e.g. anti-NGF30, anti-AchR, MRP14); (ii) previous evidence of a change in their expression during the physiological sleep/waking cycle and after short-term sleep deprivation (e.g. Arc, BDNF, Homer, BiP, VGF); (iii) their characterized function (e.g. synaptic plasticity: Arc, BDNF, Homer; response to cellular stress: MRP14, BiP; glial function, e.g. Plp, MAG, Mog).


EEG analysis

Table 1 shows mean ± SEM values of wakefulness, non-rapid eye movement (NREM) sleep and REM sleep before death for the five experimental groups. The percentages for each behavioral state refer in all cases to the last 8 h before death, which represents the entire duration of the experiment for the first three experimental groups (S, s-SD, W). Table 1 shows that with the DOW method l-SD rats were awake ∼ 82% of the last 8 h of recording. During the entire 7-day period of sleep deprivation, l-SD animals lost on average (mean ± SEM) 65 ± 3% of their daily baseline amount of NREM sleep and 94 ± 5% of their daily baseline amount of REM sleep. Thus, the DOW was effective in enforcing wakefulness, although it could not totally prevent rats from sleeping. Importantly, in this study, as in previous studies (Rechtschaffen and Bergmann 2002), YCs could not maintain all their baseline sleep. Specifically, in this experiment YC rats lost about one-third (30 ± 17%) of their daily baseline value of NREM sleep and about half (45 ± 28%) of their REM sleep. Thus, these control rats should more properly be considered as sleep-restricted rats. As expected (Rechtschaffen and Bergmann 2002), all l-SD rats showed classic symptoms of the sleep deprivation syndrome, including an increase in energy expenditure, disheveled appearance of fur, and lesions of paws and tail (data not shown).

Table 1.   Percentages of wakefulness, NREM sleep and REM sleep for the last 6–8 recording hours before death in the five experimental groups
   Behavioral state (%)
  1. Values are mean ± SEM percentage during last 8 h of recordings. Values for the first 3 groups were previously published in (Cirelli et al. 2004).

Spontaneous sleep (8 h)S (n = 6)Asleep23.4 ± 1.460.7 ±1.215.9 ± 0.7
Short-term sleep deprivation (8 h)s-SD (n = 6)Acutely sleep deprived95.9 ± 0.44.1 ± 0.40.0 ± 0.0
Spontaneous wakefulness (8 h)W (n = 6)Awake75.8 ± 3.220.3 ± 2.63.9 ± 0.8
Long-term sleep deprivation (7 days)l-SD (n = 7)Chronically sleep deprived81.7 ± 1.917.5 ± 2.10.8 ± 0.2
Long-term sleep restriction (7 days)YC (n = 7)Sleep restricted (yoked to l-SD)61.5 ± 1.233.5 ± 0.85.0 ± 0.7

Transcripts with increased expression after 7 days of sleep deprivation

We define transcripts ‘up in long-term sleep deprivation’ as those whose expression increased in long-term sleep deprivation relative to both acute waking conditions: short-term sleep deprivation and spontaneous wakefulness. To identify these transcripts we performed a conjunction analysis, i.e. we compared l-SD rats with s-SD rats, and l-SD rats with W rats, and then considered only the genes up-regulated in l-SD rats in both comparisons. The s-SD and W rats were continuously awake for about 8 h, but at a different circadian time (∼ 17.00 and ∼ 05.00 hours), and wakefulness was maintained with (novel objects in s-SD) or without (W) external stimulation. The l-SD rats, on the other hand, were forcibly kept awake for 7 days by the DOW method, and were killed at ∼ 11.00 hours. The use of two acute waking conditions was aimed at limiting confounding effects due to circadian time, the presence or absence of an external stimulus to maintain wakefulness, and the nature of that stimulus. This analysis identified 75 transcripts, including 23 ESTs, which are listed in Table 2 and Table S1. These ‘up in l-SD’ transcripts, however, do not necessarily represent only those genes whose expression is specifically sensitive to chronic sleep loss, because l-SD rats were subjected to the disk stimulation, whereas s-SD and W rats were not.

To distinguish the effects of chronic sleep loss from those related to the DOW method per se we performed additional comparisons, all listed in the tables (columns 6–9) and summarized in Fig. 1. First, we took advantage of the YCs, and determined which genes were up-regulated in the l-SD versus YC comparison. The 24 genes thus identified (Fig. 1, A) must be up-regulated as a result of the extreme sleep loss in l-SD rats, given that both l-SD and YC animals were identically exposed to the DOW. They included the transcripts coding for the transcription factor Neurod1, several immunoglobulins, the stress response protein MRP14, globins and minoxidil sulfotransferase.

Figure 1.

 Schematic showing the transcripts up- and down-regulated after long-term sleep deprivation (l-SD vs. s-SD and W). A–D indicates the four steps of conjunction analysis used to distinguish the role of chronic sleep loss and disk stimulation. For details, see text.

Next, we considered whether or not the genes up-regulated in l-SD rats were also up-regulated in YC rats compared with both s-SD and W rats. The 34 genes up-regulated in l-SD but not in YC rats must also be induced by extreme sleep loss rather than by the DOW (Figs 1, B). They included those coding for the transcription factors Atf5 and c-myc, the stress response proteins heat-shock protein (Hsp27) and α-B-crystallin (cryab), and the neuropeptide cortistatin.

We then considered which, among the transcripts up-regulated in both l-SD rats and YC rats, were previously identified as ‘waking’ genes (i.e. up in s-SD + W rats relative to S rats; Cirelli et al. 2004). We found only one transcript, coding for the Hsp70-1a. This transcript is probably induced by sleep loss – extreme sleep deprivation in l-SD rats and chronic sleep restriction in YC rats – rather than purely by the DOW, as it was also up-regulated in the acute waking conditions that did not involve the DOW (Fig. 1, C).

Finally, we identified four transcripts (including two ESTs) that were up-regulated in both l-SD and YC rats, which we had previously characterized as ‘sleep’ genes (i.e. up in S rats relative to s-SD + W rats; Cirelli et al. 2004) (Fig. 1, D). Two of these transcripts code for the D site albumin promoter binding protein (Dbp) and squalene synthetase. This left 12 of 75 transcripts with increased expression in l-SD rats that were presumably up-regulated because of non-specific DOW effects. Among these were the transcripts coding for the zinc finger protein RIZ, angiotensinogen and neuropeptide Y.

As mentioned in the methods section, microarray results were validated using qPCR on independent groups of animals. We evaluated the expression of 16 transcripts identified by the microarray analysis as ‘not changing’, ‘increased in l-SD’ or ‘decreased in l-SD’. In all but one case qPCR confirmed the array data (see Tables 2 and 3, and Tables S1–S4). The exception was the transcript for MAG, which according to the array analysis was decreased in s-SD rats but not in l-SD rats relative to S rats (see below), whereas the opposite was found with qPCR. This transcript was therefore not included in Table S4.

Transcripts with decreased expression after 7 days of sleep deprivation

Table 3 (see also Table S2) lists the ‘down in l-SD’ transcripts, whose expression decreased in long-term sleep deprivation relative to both short-term sleep deprivation and spontaneous wakefulness. We identified 28 transcripts, including 10 ESTs. To distinguish the effects of chronic sleep loss from those of DOW stimulation, we performed the same additional comparisons as discussed above for the up-regulated transcripts. For 16 of the 28 transcripts the down-regulation in l-SD rats was related to chronic sleep loss and not to DOW effects. They include those for the type I procollagen, dihydrolipoamide acetyltransferase, Cyp26b1, and several ESTs. Eight of these transcripts were also previously identified as waking genes, including those coding for three immediate early genes (Homer1, Egr2, Tieg) and pyruvate dehydrogenase phosphatase 1.

Comparisons between short-term and long-term sleep deprivation

To determine to what extent short-term and long-term sleep loss affect the expression of the same transcripts we separately compared S rats with s-SD rats and with l-SD rats, and examined the overlap between the differentially expressed genes. Tables S3 and S4 list the transcripts significantly up- or down-regulated in both sleep deprivation conditions relative to sleep, as well as those specifically changed relative to sleep in each sleep deprivation group. It is evident from these tables that the molecular correlates of acute and chronic sleep loss are not identical, and that more genes change their expression after long-term than after short-term sleep deprivation, although we cannot rule out the possibility that some of the differences may be due to the fact that l-SD rats were not killed at the same circadian time as s-SD and S rats. This is summarized in Fig. 2, which shows that only 66 genes were up-regulated in both short-term and long-term sleep loss relative to sleep. These genes represent 49% (66 of 135) of all genes up-regulated after acute sleep deprivation relative to sleep, and only 29% (66 of 226) of those up-regulated after chronic sleep deprivation. Similarly, only 16 genes were down-regulated in both short-term and long-term sleep loss relative to sleep, representing 22% (16 of 73) of all the genes down-regulated after acute sleep deprivation relative to sleep, and 14% (16 of 113) of those down-regulated after chronic sleep loss. Even within the gene categories generally up- or down-regulated in both sleep deprivation conditions there were important differences. For instance, as shown in Fig. 2, a large group of genes up-regulated after both acute and chronic sleep loss included immediate early genes and plasticity-related genes. Several of them (c-fos, junB, NGFI-A, NGFI-B, NGFI-C, Egr2 and Arc) were significantly increased in both sleep deprivation conditions relative to sleep, whereas others (BDNF, Homer, Narp and CREM) showed a strong increase in expression after acute sleep loss but much less so or only a non-significant trend after chronic sleep loss. Similarly, one of the largest categories of genes down-regulated after both acute and chronic sleep loss included glial genes. In some cases, the decrease in expression was similar after acute and chronic sleep deprivation (e.g. CnpII), in other instances it was more pronounced after acute sleep loss (e.g. CAII), and in other cases it was only significant (Fabp7) after chronic sleep loss. Moreover, for some genes represented in the arrays by more than one probe set (e.g. Plp and Mog), one transcript indicated similar down-regulation, relative to sleep, in both sleep deprivation conditions, whereas another transcript showed more pronounced decreases in long-term sleep loss. For glial genes Plp and Mog the latter result was confirmed by qPCR (Table S4).

Figure 2.

 Venn diagrams of the transcripts differentially expressed between 8 h of sleep deprivation (s-SD) and sleep (S), 7 days of sleep deprivation (l-SD) and S, and of their overlap. Line graphs show normalized intensity levels (log scale) as indicated by microarray analysis for several transcripts in each category. Gene symbols are indicated in Tables 2 and 3 and in Supplementary Tables 1–4.

RPA experiments confirmed the array results for Arc. Specifically, Fig. 3(a) shows Arc mRNA levels in a pool of S and s-SD rats (n = 4/group) and in four individual l-SD rats. In agreement with the array results, Arc expression increased significantly, relative to sleep, after both short-term and long-term sleep loss. In three of the four L-SD rats, however, the increase was not as significant as in the s-SD pool, suggesting individual differences (Fig. 3a). RPA experiments also confirmed the array results for BDNF; relative to sleep, BDNF mRNA levels increased only (D10938, S76758, X67108) or more significantly (AI030286) in s-SD rats than in l-SD rats (Fig. 3b). RPA also confirmed the array results for VGF, showing that VGF expression was similarly increased in s-SD (by 30 ± 10%) and l-SD (40 ± 8%) relative to sleep. Finally, RPA was also used to validate the array results for BiP in the cerebral cortex, and to assess its expression in peripheral tissues. Figure 4 shows that RPA confirmed that cortical BiP expression was higher in all waking conditions compared with sleep, and that the most significant increase occurred after short-term sleep deprivation. This pattern is unique to the brain, as BiP mRNA levels in skeletal muscle increased after both short-term and long-term sleep loss, whereas in the liver they increased in both l-SD and YC rats.

Figure 3.

  (a) Differential expression of Arc as measured by Rnase protection assay in the cerebral cortex (cx) of rats killed after 8 h of sleep (S), 8 h of sleep deprivation (s-SD), 1 week of sleep deprivation (l-SD1–4) and YCs. A riboprobe specific for Arc was hybridized with an equal amount of total RNA from S, s-SD and YC rats (pools of four rats), and from four individual l-SD rats. Each sample was run in duplicate, and intensity levels were measured by scanning the gel with a phosphorimager. Values are mean ± SEM. (b) Differential expression of BDNF as measured by Rnase protection assay in the cerebral cortex of rats killed after 8 h of sleep (S), 8 h of sleep deprivation (s-SD), 3 h of sleep deprivation (3hSD), 1 week of sleep deprivation (l-SD) and YCs. A riboprobe specific for BDNF was hybridized with an equal amount of total pooled RNA from four to six rats for each group. Each sample was run in triplicate, and intensity levels were measured by scanning the gel with a phosphorimager. Values are mean ± SEM. The two gels were run on different days, but they used input RNA from the same S group. A β-actin antisense riboprobe was used to normalize the amount of sample RNA.

Figure 4.

 Differential expression of BiP as measured by Rnase protection assay in the cerebral cortex (cx), skeletal (sk.) muscle and liver of rats killed after 8 h of sleep (S), 8 h of sleep deprivation (s-SD), 8 h of spontaneous wakefulness (W), 1 week of sleep deprivation (l-SD) and YCs. A riboprobe specific for BiP was hybridized with an equal amount of total RNA from four to six rats for each group. Samples were run in duplicate or triplicate, and intensity levels were measured by scanning the gel with a phosphorimager. Values are mean ± SEM. A β-actin antisense riboprobe was used to normalize the amount of sample RNA. The line graph in the left panel shows normalized intensity levels (log scale) for Bip in the cerebral cortex as measured by microarray analysis in independent groups of animals.


In this study, as in all previous array studies that aimed to identify transcripts affected by sleep and wakefulness (Cirelli et al. 2004; Terao et al. 2005), arrays were hybridized with pooled RNA. Although pooling is required to reduce the cost of the experiment, it is important to assess for biological variability using independent methods. To this end, we used qPCR and RPA in independent groups of animals not used in the array analysis. In our previous rat (Cirelli et al. 2004) and fly (Cirelli et al. 2005a) studies, PCR confirmed ∼ 80% of the array results, suggesting that less than 20% of the genes identified as differentially expressed in the arrays were false positives. In this study, all but one of the 16 transcripts identified by the array analysis as not changing, increased or decreased, were confirmed using PCR. Similarly, the four genes tested using RPA were confirmed as changing in the same direction as indicated by the arrays. The higher success rate may be partly due to the bigger fold change of some of these transcripts, especially those up-regulated in l-SD rats. However, we previously found that high expression levels (i.e. a reliable detection), rather than fold changes as indicated by the array analysis, are the best predictors of whether or not the results for a gene will be confirmed by another method. In this study, up to eight array replicas were used for the l-SD and YC groups and, most importantly, transcripts were identified using a restrictive conjunction analysis. These stringent criteria may explain both the high confirmation rate, as well as the short list of transcripts.

The largest category of genes specifically up-regulated in the cerebral cortex because of chronic sleep loss codes for immunoglobulins, in most cases κ chains. This group also includes two autoantibodies, against AchR and NGF. These genes were not previously identified after short-term sleep deprivation. Moreover, all but one were not significantly up-regulated in YCs relative to sleeping animals, and in all cases were significantly up-regulated in l-SD rats relative to their YCs. Thus, it seems that the up-regulation of these immunoglobulins requires a sleep deprivation that is both chronic and extreme, and thus cannot be triggered by 8 h of sleep loss or by chronic sleep restriction.

It has been suggested that sleep and sleep deprivation are associated with changes in immune function. Whether sleep loss actually results in an impairment of immune functions, however, remains unclear (Benca et al. 1997; Bryant et al. 2004). Both animal and human studies have shown that prolonged sleep deprivation results in activation of the immune response. In humans, 3 days of sleep deprivation produce increases in natural killer cell activity and in granulocyte and monocyte counts (with no change in lymphocytes) (Dinges et al. 1994). In a recent study Everson found that rats deprived of sleep for up to 20 days with the DOW show leukocytosis with increased counts of neutrophils and monocytes (and a trend toward decreased counts of lymphocytes), induction of pro-inflammatory cytokines and chemokines, and increased production of serum IgM, IgG and IgA, consistent with polyclonal activation of B lymphocytes (Everson 2005). The author suggests that this broad-based antibody production may be in response to the endotoxemia and the presence of opportunistic pathogens in internal tissues of rats sleep deprived with the DOW, both of which are detected at an early stage of sleep deprivation (Everson and Toth 2000). The present study and the Everson study used the DOW, and this method is currently the only one available to enforce prolonged sleep loss in animals with an adequate stimulation control. Thus, until novel methods are designed, it may be difficult to establish whether chronic sleep deprivation in animals always results in increased expression of immunoglobulins. In humans, very few studies have analyzed the effects of sleep deprivation on immunoglobulin levels, and with contrasting results. Boyum et al. (1996) found a decrease in Ig levels in subjects enrolled in 5–7 days of a military training course, but in that study sleep restriction was associated with continuous physical exercise and caloric restriction. Ozturk et al. (1999) found no change in serum IgG and IgM, but their subjects were deprived of sleep for only 48 h.

In a recent study in rats subjected to long-term sleep deprivation Everson mentioned that polyclonal B responses, such those associated with chronic antigenic stimulation, have the potential to induce autoimmune response (Everson 2005). Interestingly, we found in this study that the expression of two autoantibodies was increased after 1 week of sleep deprivation. They are directed against NGF and the AchR. Anti-NGF antibodies, when injected into rat cerebral cortex, can produce degeneration of cortical cholinergic boutons (Debeir et al. 1999) and disrupt learning (Gutierrez et al. 1997). Moreover, transgenic mice overexpressing anti-NGF antibodies develop an age-dependent neurodegenerative pathology with dementia-like symptoms (Capsoni et al. 2000). AchR antibodies, on the other hand, when infused in the rat cerebral cortex, produce fatigue and ataxia (e.g. Gomez et al. 1984). Thus, it cannot be excluded that these antibodies may contribute to the sleepiness, decreased attention and fatigue associated with chronic sleep deprivation. Autoantibodies have also been described in a few cases of severe insomnia, but never as a consequence of relatively acute sleep deprivation. Liguori et al. (2001) described a patient with Morvan's syndrome whose severe insomnia, associated with the presence of autoantibodies against voltage-dependent potassium channels, improved after plasma exchange. (Batocchi et al. (2001) reported a case of multiple cranial nerve palsy, recurrent episodes of total insomnia, and respiratory crises that responded to plasma exchange and immunosuppressive treatment. In that case the serum was negative for antibodies against voltage-dependent potassium channels, but positive for antibodies against GABAergic synapses. In the case described by Liguori et al. it is possible that the presence of the autoantibodies against potassium channels was at least partially responsible for the insomnia (Cirelli et al. 2005b). However, the autoantibodies were associated with a paraneoplastic syndrome, and thus it remains unclear whether prolonged sleep loss per se can trigger an autoimmune response.

Among other genes up-regulated by chronic sleep loss there were several stress response genes. One codes for the MRP14 (also called S100 calcium-binding protein A9 or calgranulin B). MRP14 is detected on microglial cells in bacterial encephalitis, Alzheimer's disease and after cerebral infarction (Postler et al. 1997;Staba et al. 2002). Others code for the small heat-shock proteins Hsp27 (Hspb1) and Cryab. These genes are strongly induced in glial cells (astrocytes and oligodendrocytes) after different forms of stress, and their induction may have a protective effect (Goldbaum and Richter-Landsberg 2001). Among the stress response genes not up-regulated in l-SD rats is BiP. BiP is the major chaperone of the endoplasmic reticulum and is involved in the degradation of misfolded proteins. BiP mRNA levels increase during short-term sleep deprivation in the brain of rats (Cirelli et al. 2004), hamsters (Cirelli C., DeBoer T., Tobler I., unpublished results) and sparrows (Jones S., Benca R., Cirelli C., unpublished results). In mouse cerebral cortex BiP expression is increased by as little as 6 h of sleep deprivation, which also causes a decrease in protein synthesis, another sign of the cellular stress response called the unfolded protein response (Naidoo et al. 2005). Our array analyses, confirmed by RPA experiments, showed that in the cerebral cortex BiP mRNA levels do not increase after long-term sleep deprivation as much as after short-term sleep deprivation. Interestingly, RPA experiments also showed that BiP expression is similarly induced after short-term and long-term sleep deprivation in skeletal muscle, whereas in the liver the most significant induction occurs in animals housed in the DOW apparatus, including YCs. Thus, BiP mRNA levels do not seem to reflect the duration of sleep loss.

Other genes specifically up-regulated in long-term sleep deprived rats code for minoxidil sulfotransferase, the α and β chains of hemoglobin, and cortistatin. These genes are also induced after short periods of spontaneous or forced wakefulness (Cirelli et al. 2004; Spier and de Lecea 2000), but more so after prolonged sleep deprivation, and more significantly in l-SD rats than in their YCs. The minoxidil (aryl) sulfotranserase belongs to the phenol sulfotransferase family and was originally identified in the liver (Hirshey et al. 1992). It is a class II sulfoconjugation enzyme involved in the catabolism of catecholamines and in the detoxification of drugs (Yeh and Yen 2005). In humans, two forms of phenol sulfotransferases have been described, a phenol-preferring form and a monoamine-preferring form; work in our laboratory is currently characterizing the substrate specificity of the enzyme induced by sleep deprivation. Globins are a family of heme proteins that can bind, transport and scavenge O2, CO and NO. RNAs for both the α and β chains of hemoglobin cycle robustly in the SCN, with peak expression at the end of the dark period. In mouse brain, the same genes are also strongly induced by a light pulse given at night (Ben-Shlomo et al. 2005). In our previous study, we found that these genes are induced in the rat cerebral cortex during spontaneous wakefulness at night, but also after short-term sleep deprivation during the day, suggesting that behavioral state per se can induce their expression independently of circadian time. Because these genes are expressed at a higher level after long-term than after short-term sleep deprivation, their induction is unlikely to reflect the energy demand of waking per se, because prolonged sleep deprivation in animals is accompanied by a decrease, rather than an increase, in cerebral glucose utilization (Everson et al. 1994). Moreover, in Djungarian hamster cerebral cortex they are up-regulated, relative to sleep, during short-term sleep deprivation as well as during torpor, when brain metabolism is reduced (DeBoer T., Tobler I., Cirelli C., manuscript in preparation). Thus, the up-regulation of globin RNAs may be more directly related to their role as extracellular scavengers of NO and CO, and thus be part of a cellular stress response. Cortistatin is a neuropeptide mainly expressed in GABAergic interneurons of the cerebral cortex; when infused into rat brain ventricles it increases the time spent in slow wave sleep (Spier and de Lecea 2000). Its pronounced up-regulation in long-term sleep deprivation may reflect increased sleep pressure.

Intriguingly, two transcripts up-regulated in l-SD rats code for Dbp and squalene synthase, which are also up-regulated during spontaneous sleep. Among the factors that could control the expression of these genes in both conditions, one intriguing possibility is that Dbp and squalene synthase expression somehow reflects a decrease in brain metabolism, a condition shared by both sleep (Maquet 1995) and long-term sleep deprivation (Everson et al. 1994).

Twenty-eight transcripts were down-regulated in l-SD rats. One transcript specifically down-regulated because of chronic sleep loss codes for procollagen type I. Three more transcripts coding for procollagen type I were also down-regulated in l-SD rats, but they also showed decreased expression in YC rats, and did not show a further decrease in l-SD rats relative to YC rats. Thus, it remains unclear whether changes in collagen expression are related to DOW effects or to chronic sleep loss. Another transcript down-regulated because of chronic sleep loss codes for dihydrolipoamide acetyltransferase. This enzyme is a component of the pyruvate dehydrogenase complex involved in the synthesis of acetyl-CoA, and therefore its decreased expression is consistent with the decrease in brain metabolism in l-SD rats.

Many of the stress response genes up-regulated after prolonged sleep loss relative to short-term sleep deprivation and spontaneous wakefulness are preferentially expressed in glial cells, either in microglia (MRP14) or in astrocytes and oligodendrocytes (Hsp27, metallothionein 1–2). On the other hand, several genes down-regulated after both short-term and long-term sleep deprivation relative to sleep are also expressed in glial cells (this paper; Cirelli et al. 2004). In this study, for instance, we confirmed with qPCR that long-term sleep deprivation down-regulated the expression of two myelin-related genes, one coding for plasmolipin, which constitutes 50% of myelin protein, and the other for CD9, a membrane protein normally expressed in the mature myelin sheath and a gene up-regulated during sleep in the cerebellum (Cirelli et al. 2004). Overall, these findings suggest that sustained sleep loss may trigger a generalized inflammatory and stress response in the brain. Glial cells may help to protect neurons against this cellular insult, but may also suffer some of its negative consequences. Future studies will determine whether sleep loss may be detrimental to the maintenance of cellular membranes and more specifically to myelin.


This work was funded by the National Institute of Mental Health (R01 MH65135). We thank Anne Luebke for technical assistance.