Journal of Neurophysiology

Tumor Necrosis Factor (TNF) Ligand and TNF Receptor Deficiency Affects Sleep and the Sleep EEG

Tom Deboer, Adriano Fontana, Irene Tobler

Abstract

Tumor necrosis factor (TNF) and lymphotoxin-α (LT-α) are proinflammatory cytokines involved in host defense and pathogenesis of various diseases. In addition, there is evidence that TNF is involved in sleep. TNF and LT-α both bind to the tumor necrosis factor receptors (TNFR). Recently, it was shown that TNF receptor 1 (TNFR1) knockout mice (R1KO) sleep less during the light period than controls. We investigated the effect of a TNF and LT-α double deficiency on sleep in mice (Ligand KO) and compared their sleep with that of R1KO, TNFR2 knockout (R2KO) mice, and wild-type (WT) controls. All mice were adapted to a 12:12 h light:dark cycle and their electroencephalographs (EEG) and electromyographs (EMG) were continuously recorded during a baseline day, 6-h sleep deprivation (SD), and 18-h recovery. Ligand KO and R2KO had 15% less rapid eye movement (REM) sleep during the baseline light period due to a reduction in REM sleep episode frequency. After SD, all genotypes showed an initial increase in slow-wave activity (SWA) (EEG power density between 0.75 and 4.0 Hz) in non-REM sleep, which gradually declined in the following hours. In Ligand KO the increase was mainly caused by an increase in fast SWA (2.75–4.0 Hz), which was also increased in R2KO. In contrast, in R1KO mice the increase was limited to the slow portion of SWA (0.75–2.5 Hz). R2KO and WT mice showed increases in both frequency ranges. The sub-division into fast and slow SWA frequencies corresponds to previous electrophysiological data where the two types of slow-waves were induced by either excitatory or inhibitory stimuli. Our data suggest that in Ligand KO the SWA increase is caused by an increase in excitatory input to the cortex, whereas in R1KO this input seems to be almost absent.

INTRODUCTION

Tumor necrosis factor (TNF) is produced by macrophages and circulating monocytes and is a pro-inflammatory cytokine involved in host defense and pathogenesis of various diseases (Aggarwal and Vilcek 1992;Beutler 1992). In the CNS TNF is produced in astrocytes (Lieberman et al. 1989) and microglial macrophages (Frei et al. 1987) and influences the peripheral immune system through the lymphatic pathway (Dickstein et al. 1999). There is an extensive literature suggesting that TNF and other cytokines may also be involved in sleep regulation. Thus it was shown in several mammalian species that administration of TNF increases nonrapid eye movement (NREM) sleep (Dickstein et al. 1999; Fang et al. 1997; Kapas et al. 1992; Shoham et al. 1987; Terao et al. 1998). Moreover, inhibition of endogenous TNF after administration of antibodies or soluble receptors inhibits spontaneous sleep (Takahashi et al. 1995a,b) and attenuates the increase in sleep after sleep deprivation (SD) (Takahashi et al. 1996). Hypothalamic levels of TNF mRNA (Bredow et al. 1997) and protein (Floyd and Krueger 1997) are increased during the light period in rats, i.e., during the main sleep period of the animals. These and other findings led to the notion that TNF is involved in sleep regulation (reviewed by Krueger et al. 1999).

Two cell-surface receptors for TNF, TNF receptor 1 and 2 (TNFR1 and TNFR2) (55 and 75 kDa in size, respectively) have been characterized (Hohmann et al. 1989, 1990; Schall et al. 1990). The two receptors mediate distinct actions (Schutze et al. 1994; Vilcek and Lee 1991) but cannot be differentiated by pharmacological means. It is known that lymphotoxin-α (LT-α) binds to the same receptors as TNF, and consequently, administration of TNF and LT-α leads to similar biological responses in vivo and in vitro (Beutler and Van Huffel 1994). This redundancy in receptor binding and their overlapping expression pattern renders a distinction between the function of TNF and LT-α rather difficult. For this purpose, TNF receptor-deficient mice were generated, which should enable the dissection of the immunological function of the receptors (Erickson et al. 1994; Pfeffer et al. 1993; Rothe et al. 1993; Ruby et al. 1997). In addition, it was shown that TNFR1 knockout (R1KO) mice show less sleep during the light period and do not increase sleep after application of exogenous TNF (Fang et al. 1997), indicating that TNF affects sleep via TNFR1.

An alternative approach to investigate the role of the TNF receptors is by removal of the ligands, TNF and LT-α. The consequences of the combined deficiency of TNF and LT-α on the immune system were investigated recently by generating TNF and LT-α double-deficient mice (Amiot et al. 1997; Eugster et al. 1996). Our aim was to investigate the role of TNF receptors in sleep and sleep regulation in these mice (Ligand KO). On the basis of the results with R1KO mice, we expected that the Ligand KO mice would show less sleep during the light period. To enhance physiological sleep pressure, a sleep deprivation (SD) was performed and sleep and sleep regulation of the double-deficient mice were compared with that of wild-type control mice (WT) and R1KO and TNFR2 knockout (R2KO) mice.

METHODS

Animals

Adult mice of the four genotypes were used [WT,n = 12; TNF and lymphotoxin-α double deficient (Ligand KO), n = 9; R1KO, n = 8; R2KO,n = 8]. The KO mice were generated on 129/SV background and backcrossed for 6 (Ligand KO) or 10 generations (R1KO and R2KO) with C57BL/6. Normal C57BL/6 mice were used as WT controls. The animals were maintained on a 12:12 h light:dark schedule (lights on from 0800–2000 h; daylight-type fluorescent tubes, 18 W, 50–100 lux at the level of the mice), individually kept in Macrolon cages (36 × 20 × 35 cm), and placed in sound-attenuated chambers. Food and water were available ad libitum. The animals were adapted for a minimum of 3 wk to these conditions.

Surgery

The mice were implanted with electroencephalographic (EEG) and electromyographic (EMG) electrodes under deep pentobarbital sodium anesthesia (Nembutal sodium, 80 mg/kg ip, volume approximately 0.5 ml). Two gold-plated miniature screws (Ø0.9 mm), placed over the right occipital cortex (2–3 mm lateral to midline, 2 mm posterior to bregma) and the cerebellum (at midline, 1 mm posterior to lambda), served as epidural EEG electrodes. The EMG was recorded with two gold wires (Ø0.2 mm) inserted into the neck muscles. The electrodes were connected to stainless steel wires that were glued to the skull with dental cement. At least 3 wk were allowed for recovery. Age at recording onset was approximately 16 wk and did not differ between genotypes.

Experimental protocol

A 24-h baseline recording starting at lights on preceded the 6-h SD. Mice were also recorded during the SD and for the following 18 h. SD began at light onset and was performed by introducing objects (e.g., nesting material) into the cage, and later by tapping on the cage whenever the animal appeared drowsy or the EEG exhibited slow-waves. Halfway through the SD, the mice were provided with new cages, which induced additional stimulation and elicited exploratory behavior. To minimize stress, some material from the old cage was transferred to the new one. The mice were never disturbed during feeding and drinking.

The EEG and EMG signals were amplified (amplification factor ∼2000), conditioned by analog filters (high-pass filter: −3 dB at 0.016 Hz; low-pass filter: −3 dB at 40 Hz, less than −35 dB at 128 Hz), sampled with 512 Hz, digitally filtered (EEG: low-pass FIR filter, 25 Hz; EMG: band-pass FIR filter, 20–50 Hz), and stored with a resolution of 128 Hz. EEG power spectra were computed for consecutive 4-s epochs by a fast Fourier transform (FFT) routine within the frequency range of 0.25–25.0 Hz. Between 0.25 and 5.0 Hz, the values were collapsed in 0.5-Hz bins and between 5.25 and 25.0 Hz the values were collapsed into 1-Hz bins. EMG signals were integrated over 4 s and ambient temperature inside the cage was recorded at 4-s intervals. All data were recorded simultaneously and stored on optical disk. The EEG and EMG channels were calibrated by recording a 10-Hz sine-wave, 300 μVPP signal before each recording.

Analysis

Vigilance states were determined for 4-s epochs as described previously (Tobler et al. 1997). In short, waking was scored when there was a high-EMG and low-EEG amplitude and high-θ activity (EEG power density in the θ-band, 6.25–9.0 Hz), concomitant with irregular, high EMG values, and NREM sleep when there was a low-EMG and higher EEG amplitude compared with waking and high SWA, rapid eye movement (REM) sleep when the EMG and EEG amplitude was low, and high-θ activity was visible in the EEG. Epochs containing EEG artifacts were marked and excluded from the spectral analysis (12.0 ± 2.6% of recording time; >85% of all EEG artifacts occurred in waking), but vigilance states could always be determined.

The duration and frequency of NREM sleep, REM sleep, and waking episodes were determined according to criteria described previously (Deboer et al. 1994; Deboer and Tobler 1996). The algorithm is based on the frequency and duration of interruptions of the vigilance state episodes.

Overall effects within a strain were analyzed by two-way analysis of variance (ANOVA) for repeated measures, with factors “time” and “condition,” with the exception of spectral data where, due to intervals with low amount of sleep, resulting in missing values, a two-way ANOVA with factor time and condition was used. Differences between the four genotypes were analyzed by two-way ANOVA with factors “genotype” and time. Contrasts between WT and single KO groups were tested by post-hoc two-tailed t-tests when ANOVA reached significance.

RESULTS

Vigilance states

Ligand KO and R2KO mice had 15% less REM sleep during the baseline light period than controls (Table1). There were no differences in the amount of sleep between R1KO and WT mice. The lower amount of REM sleep in the Ligand KO and R2KO mice was reflected also in less REM sleep/total sleep time (TST) during the light period (Table 1). The reduced amount of REM sleep was mainly caused by a reduction in REM sleep episode frequency, which was not compensated by an increase in REM sleep episode duration (Table 2). Also, NREM sleep episode frequency was reduced in Ligand KO and R2KO and was compensated by an increase in NREM sleep episode duration. During the dark period, REM sleep episode frequency was reduced in all KO mice, compared with WT, and was compensated by an increase in REM sleep episode duration that reached the same duration as in the light period (Table 2).

View this table:
Table 1.

Vigilance states in the light and dark period and 24-h values

View this table:
Table 2.

Vigilance states episode duration and frequency in the light and dark period and 24-h values

The time course of the vigilance states in the course of the entire experiment is illustrated in Fig. 1. During the first 4–6 h of the baseline light period, all KO mice had less REM sleep than WT, which reached significance in the Ligand KO and R2KO mice only (Fig. 1). The time course of waking and NREM sleep was indistinguishable between the four genotypes during baseline, and no differences were obtained in the response to SD. The amount of sleep during SD was negligible and did not differ between the genotypes (WT: 0.7 ± 0.2% of 6 h; Ligand KO: 1.1 ± 0.2; R1KO: 1.0 ± 0.3; R2KO: 1.6 ± 0.5; F = 1.594,P = 0.209 ANOVA factor “group”). REM sleep showed a prolonged increase after SD in all genotypes. NREM sleep was enhanced in the dark period only, except in R2KO mice where NREM sleep was increased immediately. Despite these differences, there was no significant difference in the effects of SD on NREM sleep and waking when the genotypes were compared.

Fig. 1.

Time course of vigilance states and slow-wave activity (SWA) [mean electroencephalographic (EEG) power density between 0.75–4.0 Hz] in non-rapid eye movement (NREM) sleep for baseline, the 6-h sleep deprivation (SD), and 18-h recovery for the 4 genotypes (wild-type: WT,n = 12; double knockout: Ligand KO,n = 9; Receptor 1 KO: R1KO, n = 8; Receptor 2 KO: R2KO, n = 8; each plot shows WT in circles). Curves connect mean 2-h values (means ± SE) of NREM sleep, REM sleep, waking, and SWA. SWA is expressed relative to the mean 24-h baseline value (=100%). The vertical lines and black and white bars indicate the light:dark cycle. Significant differences between baseline and recovery within a genotype are indicated by open (WT) and filled triangles [KO; P < 0.05, two-tailed paired t-test, after significant analysis of variance (ANOVA)]. Orientation of the triangles indicates the direction of deviation from baseline. Asterisks indicate significant differences in REM sleep between Ligand KO or R2KO and WT mice (P < 0.05, two-tailed t-test after significant ANOVA factor “genotype”).

EEG power spectra

Ligand KO mice had higher absolute EEG power density in NREM sleep within the slow-wave range than the other genotypes. Thus power density between 0.75 and 2.5 Hz was higher in Ligand KO mice than WT and R1KO (Fig. 2) and higher in the 1.5-Hz bin than R2KO. There were no differences between genotypes in the waking and REM sleep EEG spectra (data not shown). After SD, all mice showed a similar increase in SWA (mean EEG power density 0.75–4.0 Hz) in NREM sleep (Fig. 1). However, there were distinct differences between the genotypes within the SWA band (Fig. 3). Thus the increase in Ligand KO and R2KO mice was mainly in the range of 2.75–4.0 Hz, while in WT and R1KO it was most prominent between 0.75 and 2.5 Hz (Fig. 3, top panels). During the recovery light period, this frequency range remained above baseline (BL) in Ligand KO and R2KO and above WT in Ligand KO (Fig. 3, top panels). Moreover, in the dark period power density in this frequency range was above WT levels in Ligand KO and R2KO in most of the 3-h intervals (Fig. 3, bottom panels). WT, R1KO, and R2KO showed a negative rebound in the range between 0.75 and 3.0 Hz in the dark period.

Fig. 2.

Spectral distribution of EEG power density in NREM sleep in the 4 genotypes (WT, n = 12; Ligand KO,n = 9; R1KO, n = 8; R2KO,n = 8). Between 0.25 and 5.0 Hz, values were calculated in 0.5-Hz bins and between 5.25 and 25.0 Hz in 1-Hz bins. Values are plotted at the upper limit of each bin. The curves connect logarithmic values of absolute power densities in NREM sleep (log μV2/Hz). Lines above the abscissa indicate significant differences between Ligand KO (L) and each of the other three genotypes (P < 0.05, two-tailed t-test, after significant ANOVA factor genotype).

Fig. 3.

Time course of EEG power density in NREM sleep in the light and dark period after sleep deprivation in the 4 genotypes (WT,n = 12; Ligand KO, n = 9; R1KO,n = 8; R2KO, n = 8). Between 0.25 and 5.0 Hz, values were calculated in 0.5-Hz bins and between 5.25 and 25.0 Hz in 1-Hz bins. Values are plotted at the upper limit of each bin. Curves connect means of relative power density for the consecutive two 3-h intervals of recovery expressed relative to the first two 3-h intervals of the light period of baseline (=100% light period) or the corresponding interval of the dark period (=100% dark period). Lines above the abscissa indicate frequencies that after SD differ significantly from baseline (BL; P < 0.05, 2-tailed paired t-test after significant ANOVA factor “day”) or WT (P < 0.05, 2-tailedt-test, after significant ANOVA factor genotype).

Also, in frequencies above 5 Hz, SD elicited changes in the EEG power spectra in NREM sleep. Ligand KO and R2KO showed an increase in almost all frequency bins between 7 and 25 Hz (Fig. 3, top panels). In contrast, WT and R1KO mice showed only a small increase above BL around 9 Hz in the second 3-h interval of the recovery light period. In this frequency range, the Ligand KO also showed an increase above WT in the first 3-h interval of recovery (Fig. 3, top panels). In the last two 3-h intervals of the dark period, R2KO showed increases above WT in several bins between 15 and 22 Hz (Fig. 3, bottom right panel).

Slow-wave activity

Because of the differences between genotypes in the slow-wave range of the NREM sleep EEG spectrum after SD (Fig. 3), and previous results in mice suggesting that there is a difference in the effect of SD on “slow” slow-waves (0.75–2.0 Hz) and “fast” slow-waves (2.75–4.0 Hz) (Huber et al. 2000b), SWA was subdivided into these two bands (Fig. 4). During baseline, both slow and fast slow-wave bands decreased in parallel during the first part of the light period in all genotypes. At the beginning of the dark period the fast slow-wave band increased above the values of the slow band and remained higher during the dark period. This increase was particularly evident in the Ligand KO, where fast SWA increased above the WT level already during the last 2-h interval of the light period (Fig. 4). After SD the relationship between the fast and slow SWA bands in WT persisted. Both bands showed an increase above baseline levels in the initial 2 h of recovery in WT and R2KO and exhibited a subsequent negative rebound in the dark period. However, in Ligand KO mice only the fast SWA band had increased above the baseline level and remained high throughout the rest of the recovery period. During the dark period, fast SWA was above levels reached by WT, and it was not followed by a negative rebound. In contrast, in R1KO only the slow SWA increased above baseline levels and showed a negative rebound during the dark period.

Fig. 4.

Time course of EEG power density in 2 frequency bands (0.75–2.5 and 2.75–4.0 Hz) in NREM sleep during the baseline day and recovery after 6-h sleep deprivation in the 4 genotypes (WT, n = 12; Ligand KO, n = 9; R1KO, n = 8; R2KO, n = 8). For each frequency band the values are expressed as a percentage of its 24-h baseline mean (=100%). Curves connect 2-h mean values (±SE). Differences between baseline and recovery within each genotype are indicated by open (0.75–2.5 Hz) and closed triangles (2.75–4.0 Hz, P < 0.05 2-tailed paired t-test after significant ANOVA factor day). Orientation of triangles indicates the direction of deviation from baseline. Asterisks indicate significant differences between Ligand KO and WT mice for the fast slow-wave activity (2.75–4.0 Hz;P < 0.05, 2-tailed t-test after significant ANOVA factor genotype).

DISCUSSION

No difference was found in the amount of sleep between the genotypes. This is in contrast with previous data where a 20% reduction in NREM sleep was reported for R1KO mice (Fang et al. 1997). Since this reduction was attributed to the deficiency of a signal coming from the TNF R1 receptor, we expected that the Ligand KO and our R1KO mice would sleep less as well. It is unlikely that in our mice compensation occurs via a different mechanism, since we did not obtain less NREM sleep in any genotype. The present results are in accordance with the interpretation that the reduction in NREM sleep reported previously may be a consequence of differences in the genetic background of the R1KO mice and WT control (Fang et al. 1997).

Since the mice do not sleep less when lacking either the ligand or the TNFR1 receptor, the data suggest that the role of TNF in the regulation of spontaneous NREM sleep is not as large as was suggested previously (Fang et al. 1997; Krueger et al. 1999). Their relation with previous data obtained with TNF antibodies and soluble TNF receptor fragments needs to be investigated further. These results were obtained in rats and rabbits (Takahashi et al. 1995a,b) and mice may react differently to this type of treatment. In mice it was also found that NREM sleep increases after application of TNF (Fang et al. 1997). However, it is known that mice often react differently to treatments than rats, e.g., mice become hypothermic, instead of producing a fever when infected with influenza (Klein et al. 1992; Toth et al. 1995). Therefore the effects of TNF antibodies or soluble TNF receptors still need to be investigated in mice.

The clear reduction in the amount of REM sleep in Ligand KO and R2KO mice contrasted the lack of difference in NREM sleep. This reduction was restricted to the light period and was due to a lower amount of NREM sleep and REM sleep episodes. The reduction in NREM sleep episodes was compensated by an increase in their duration, whereas the reduction of REM sleep episode frequency was not compensated. In the dark period all KO mice showed an increase in REM sleep episode duration to light period levels, accompanied by a reduction in REM sleep episode frequency. The absence of TNF receptors or their ligands seems to reduce the probability of initiating REM sleep, a deficiency that can only be compensated by increasing the duration of REM sleep episodes. The effects on REM sleep in the KO mice are similar to the effects of chronic melatonin administration in Djungarian hamsters (Deboer and Tobler 1997) or of small amounts of a serotonin (5-HT) 1A receptor agonist administered into the peribrachial region of the pedunculo pontine nuclei in cats (Sanford et al. 1994). Melatonin is known to increase 5-HT levels throughout the brain (reviewed in Sandyk 1995), and both melatonin and 5-HT treatments reduced REM sleep episode frequency without changing its duration. TNF is known to increase 5-HT uptake in the brain by modulating 5-HT transporter function (Mössner et al. 1998). Therefore the removal of TNF or its receptor may increase the level of 5-HT in the brain. Indeed, it was shown that in mice deficient of the TNF-α gene 5-HT metabolism is increased with enhanced 5-HT levels in several brain areas, including the medulla oblongata/pons, a brain area involved in REM sleep regulation (Yamada et al. 2000). The present data suggest that TNF influences REM sleep by changing 5-HT levels through both R1 and R2 receptors.

Previously it was shown that mice strains differ in the composition of fast and slow SWA (Franken et al. 1998; Huber et al. 2000a). After SD, both fast and slow SWA increased above baseline levels but the slow SWA increase persisted longer (Huber et al. 2000b). The present results in the WT confirmed this finding. Here we show that different frequencies within SWA were enhanced after SD between the different genotypes. The slow portion of SWA, which increased in WT, R1KO, and R2KO, corresponds to the frequency of slow cortical oscillations (<1 Hz) and thalamic clock-like oscillations (1–4 Hz) typical for sleep (Amzica and Steriade 1998). In contrast, the fast SWA band, which was increased in WT, R2KO, and the Ligand KO, corresponds to the intrinsic delta activity of cortical neurons (3–4 Hz). The latter is the only slow-wave oscillation that is induced by depolarization of the cell membrane (Amzica and Steriade 1998). Differences were also found in the higher frequency range (above 7–8 Hz), but the differences to WT were not systematic and most of the time no differences were found in this frequency range between any of the genotypes. No differences were obtained between WT and R1KO, whereas Ligand KO and R2KO showed a similar pattern in their differences to WT (Fig. 3). The results suggest that the increase in SWA after SD in the Ligand KO is mainly caused by an excitatory input to the cortex, which is enhanced above the normal input occurring during the dark period, when fast SWA is increased as well. In contrast, in R1KO this excitatory input seems to be almost absent.

The present results indicate that a similar increase in the traditional SWA band after SD does not necessarily indicate a similar reaction in the slow-wave range of the EEG spectrum. Detailed analysis of the EEG power density spectrum within SWA reveals clear differences in how this increase is achieved. Moreover, within the slow-wave range there seem to be differences in recovery speed between fast and slow SWA (Huber et al. 2000b). It has previously been shown in rats that application of interleukin-1β in some cases not only increased power density in the slow-wave range, but also in frequencies above 10 Hz and between 16 and 25 Hz (Lancel et al. 1996). Here we show that depending on genotype, either the fast or slow portions or the SWA band increases after SD. This may reflect differences in the influence of TNF receptors on distinct groups of neurons and stresses the importance of obtaining detailed spectral data.

In conclusion, the deficiency of one of two TNF receptors or the ligand to that receptor does not affect the amount of NREM sleep in mice. However, it does reduce the amount of REM sleep in a similar way as after administration of 5-HT into the brain. The increase in SWA after SD is changed in Ligand KO and R2KO mice where the increase shifted in the direction of fast slow-waves. Since these fast slow-waves are induced by an excitatory input to the cortex, the data suggest that, in Ligand KO and R2KO mice, SD causes an exaggerated excitatory input above the level normally found during the dark period.

Acknowledgments

We thank Dr. H. P. Eugster for providing the TNF ligand double-deficient mice, Dr. H. Bluethmann for providing the TNF R1 and TNF R2 KO mice, and M. Lueber for help with scoring.

This research was supported by Swiss National Science Foundation Grant 3100-053005.97.

Footnotes

  • Present address and address for reprint requests: T. de Boer, Dept. of Physiology, Leiden University Medical Centre, Postbus 9604, 2300 RC Leiden, The Netherlands (E-mail: Tom.de_Boer{at}lumc.nl).

REFERENCES

View Abstract