At the cellular level, circadian rhythms are thought to be generated by transcriptional-translational feedback loops made up of positively and negatively acting transcriptional regulators 1. In mammals, the core positive elements are CLOCK and NPAS2 and their obligate dimerization partner BMAL1. These transcription factors drive per and cry transcription. PER and CRY proteins, in turn, interact with CLOCK/NPAS2:BMAL1 heterodimers to inhibit their own transcription, thus constituting the negative feedback elements. The circuitry is clearly more complex as additional feedback loops and post-translational modifications are involved 1. Clock genes underlie circadian rhythm generation in many tissues, but the circadian rhythm in the suprachiasmatic nucleus (SCN) is required for the manifestation of overt physiological and behavioral rhythms, and is therefore considered the master circadian pacemaker 12.

Although the role of clock genes in generating circadian rhythms is firmly established, we and others have found that they also play a role in the homeostatic regulation of non-REM sleep (NREMS) 345. This notion is based on the observation that mice deficient for one or more clock genes show an altered response to sleep deprivation (SD) and on the observation that clock-gene expression in brain areas outside the SCN, notable the cerebral cortex, is influenced by the prior occurrence of sleep and wake. For instance, the typical responses in NREMS to SD were reduced or absent in mice lacking the cry1 and -2 (cry1,2^-/-) 4 or bmal1 (bmal1^-/-) 6 genes compared to wild-type controls. The diminished NREMS variables included duration, consolidation, and EEG delta power. EEG delta power quantifies the prevalence and amplitude of delta oscillations and is widely used to index the homeostatic need for NREMS 7. Sleep under undisturbed, baseline conditions was also altered in mutant mice. clock-mutant (clock^Δ/Δ) 8 and npas2^-/-3 mice spent less time in NREMS whereas cry1,2^-/-4 and bmal1^-/- 6 mice slept more. NREMS consolidation was increased in cry1,2^-/- mice while the opposite was observed in clock^Δ/Δ 8 and bmal1^-/- 6 mice and in mice lacking albumin D-binding protein (dbp^-/-)9; a gene under the transcriptional control of CLOCK 10 and that also has been implicated in circadian rhythm generation 11. Average levels of EEG delta power were reduced in clock^Δ/Δ 8, dbp^-/-9, and bmal1^-/- 6 mice and increased in cry1,2^-/- 4 mice.

Clock-gene expression in the brain, notably the cerebral cortex, depends on the sleep-wake (or rest-activity) distribution. Cortical levels of per are high at times when sleep drive is high, irrespective of the phase at which the circadian expression of per peaks in the SCN. Thus, in both nocturnal and diurnal species, per expression in the cortex is maximal in conjunction with the major waking episode 121314. Under conditions where the phase (methamphetamine administration, restricted feeding) or distribution (circadian splitting) of locomotor activity is altered, per expression in the cortex parallels the overt rhythm of wakefulness, while per expression in the SCN remains unaffected 13151617. Consistent with these observations, our data demonstrate that per expression increases when mice are kept awake 34. The lack or alteration of particular clock genes in the various mouse models mentioned above, changes the expression levels of other clock genes in the circuitry. For instance, basal per expression levels are decreased in clock^Δ/Δ mice 1018 and increased in cry1,2^-/- mice 4192021 which corresponds to their opposing sleep phenotypes (see above). Also, the SD-induced increase in per expression is reduced in npas2^-/- mice 3 which suggest that NPAS2 is implicated in linking per2 expression to the sleep-wake distribution. Likewise, npas2^-/- mice lack circadian rhythms in cortical per2 expression despite an intact circadian organization of sleep and waking 12.

The finding that NPAS2 couples wakefulness to clock gene expression is of particular interest given the evidence that NPAS2 (and CLOCK) -mediated transcription is redox sensitive 22. Redox sensitivity suggests that this clock gene network may be linked to cellular energy metabolism 23. Since restoration of an optimal neural energy state has long been considered a possible function of sleep 24, this gene network might underlie a fundamental aspect of sleep homeostasis. If so, another critical gene in this network is likely to be ldha (lactate dehydrogenase A). ldha can be induced in cell culture by NPAS2, has multiple E-box sites, is a critical enzyme for neuronal energy metabolism, and utilizes NADH as a co-factor 1222. Therefore, we decided to investigate gene expression in the forebrain in greater detail over the course of a SD and after recovery sleep for the CLOCK and NPAS2 target genes per1,per2, dbp, and ldha. Futhermore, we undertook this examination of the sleep-wake dependent changes in gene expression in three inbred strains of mice known to differ in the rate at which sleep need, quantified as EEG delta power, increases during wakefulness 25. We reasoned that if clock genes are implicated in sleep homeostasis, then the time course of changes in gene expression during a SD in these strains should vary according to the published differences in EEG delta power dynamics.

We previously found that cortical mRNA levels of the clock genes per1 and per2 were increased after SD in rats and B6 mice 4. Here we have extended that observation in two additional inbred strains; AK and D2. Furthermore, by following the time course of these changes during SD and recovery sleep, we established that per expression changes as a function of the time-spent-awake and asleep. Such a time course is consistent with a role for these genes in the homeostatic regulation of sleep since the need for sleep increases with duration of wake and decreases with duration of sleep. A novel aspect of this study is the comparison of the SD-induced changes in gene expression among three inbred strains for which the homeostatic regulation of sleep differs 25. During both spontaneous and enforced periods of wakefulness the need for sleep, indexed by the level of EEG delta power, increased fastest in AK, intermediate in B6, and slowest in D2 mice. If indeed per expression would simply parallel EEG delta power we would expect to find the largest increase in per expression after SD in AK mice and the smallest in D2. Here we found evidence for the opposite; although per expression was consistently increased with SD, this increase was modulated by genotype and was found to be highest in D2 and lowest in AK. In addition, per2 expression remained elevated during recovery sleep in D2 mice only. Thus high, sustained levels of per2 expression might negatively impact the sleep recovery process and/or expression of EEG delta oscillations during NREMS.

An implicit assumption in our interpretation of our results is that the sleep-wake dependent changes in the expression of per1, per2, and dbp are mediated by CLOCK and/or NPAS2 induced transcription. The observation in npas2^-/- mice that cortical per2 expression no longer follows the circadian sleep-wake distribution 12, together with the observations that the increase in per2 expression after SD is reduced in these mice 3, and that the homeostatic response to SD is altered in npas2^-/-, cry1,2^-/-, and bmal1^-/- mice (see Introduction) strongly support this assumption. Nevertheless, expression of the genes we examined can be induced by other transcriptional regulators as well. E.g., per1 can be readily induced by cAMP-responsive element-binding protein (CREB) 293637. CREB signaling also increases during prolonged wakefulness 3839 and thus might have contributed to the SD-dependent changes in per1 reported here. Another example is the close link between glucocorticoid signaling and per1 and per2 expression 40414243 through which SD, which is accompanied by an increase in plasma corticosterone levels, could have affected per expression. In fact, we found that the SD-induced increase in corticosterone levels was 3-fold higher in D2 mice compared to the other two strains (Franken et al., in preparation) consistent with the larger relative increase in per expression in this strain observed here. Evidence for an altered transcriptional regulation of the four genes during SD (compared to that during control conditions) is provided by the altered relationship between the expression patterns we report here.

The cortical expression of dbp was decreased after SD confirming previously published data 444. Given the fact that per1, per2, and dbp transcription are, in part, under the same transcriptional control, these opposite effects of the SD are not easily explained. PER proteins are co-repressors of CLOCK and NPAS2 induced transcription. Thus, assuming that increased per expression is followed by increased PER levels within the time frame of the SD, PER-mediated transcriptional repression is an obvious possible pathway by which dbp expression could be decreased. In line with this assertion, we found that the decrease in dbp expression to be largest in the strain with the largest increase in per expression; i.e. D2. The observation that per and dbp expression were nevertheless positively correlated during the SD is, however, inconsistent with this scenario. In addition, per expression itself could be expected to decrease as well, at least the contribution of CLOCK and NPAS2 mediated transcription to the total per mRNA. Finally, the time course of the peripheral changes in dbp and per expression under control conditions also doesn't support a dbp suppressing role for PER. dbp mRNA levels peak at ZT8 and are declining when per mRNA starts to increase 16262728.

The in situ hybridization studies indicated that SD induced pronounced changes in per expression in cerebellum and cerebral cortex. This is in line with our previous findings using qPCR showing that per expression was increased in the cerebral cortex but not in brainstem and hypothalamus 4 and with the per2 increase observed after SD both in cerebellum and cerebral cortex 44. While the in situ data are limited, we observed some of the highest expression and largest increases in per1 mRNA in the most frontal area of the cortex. EEG delta power is generally higher and its increase after SD more pronounced in frontal cortex relative to other cortical areas such as occipital regions 454647 in accordance with the concept that frontal areas in particular are affected by SD 48. In mice lacking a functional PER1 protein (per1^Δ/Δ) EEG delta power was decreased in the frontal derivation specifically and the SD-induced increase in EEG delta power was smaller in per1^Δ/Δ and per2^Δ/Δ mice also only in the frontal derivation 49.

Already clear from the present data is that not all changes in forebrain per expression can be attributed to changes in the sleep-wake distribution; expression increases during the latter part of the light period while animals remain mostly asleep. This increase precedes dark onset after which maximal per expression levels are reached when mice are the most active 1627. A major adaptive advantage provided by circadian rhythms is that it allows organisms to anticipate and prepare for predictable periodic changes in the environment 50. The anticipatory increase in per expression might be part of a molecular repertoire preparing the animal for the active period. Some types of anticipatory behavior, however, don't require an intact circadian system, as food-restriction studies have shown 51. Moreover, clock-genes in the forebrain seem to play a permissive role in such anticipatory behaviors. For example, npas2^-/- mice, which have intact circadian rhythms in overt behavior, have difficulty adapting to and anticipating altered feeding schedules 17. Besides food intake, other homeostatically regulated behaviors such as sleep seem to be similarly affected 3. Thus, the same molecular network of transcriptional regulators that sets internal time-of-day in the SCN might, in the forebrain, track and signal time without sleep (or food). Conceptually the distinction between homeostatic and circadian processes at the molecular level thus becomes unclear. Mechanistically we proposed that changes in clock-gene expression could be tied directly to cellular energy state based mainly on the observation that CLOCK and NPAS2 mediated transcription is sensitive to redox state 2223. In this respect the lack of an effect of SD on ldha expression is disappointing. Although ldha has a canonical E-box and has been shown to be activated by both NPAS2 and CLOCK in a neuroblastoma cell line 22, mRNA levels of this gene were surprisingly stable across time-of-day, experimental condition, and genetic background.

We found that the expression of three circadian clock genes changes as a function of time-spent-awake and that these changes were influenced by strain. These results further indicate that circadian clock genes play a role in the homeostatic regulation of sleep. Additional evidence in support of our hypothesis has recently been obtained in humans homozygous for one of two known polymorphisms in hPer3 52. Based on SCN lesion experiments in rats 53 and forced desynchrony studies in humans 54 it is generally thought that homeostatic and circadian processes operate separately and have different neurophysiological substrates. Our current results underscore that, at least at a molecular level, these two processes are difficult to separate. Recent data also show that at the neuro-physiological level, SD can affect the activity of SCN neurons previously thought to be dedicated exclusively to setting internal time-of-day55. Finally, the present results might be of relevance for research on the biology of mood disorders. In this field there is increasing interest in the relation between clock-genes and e.g. seasonal affective disorders 56 or bipolar disorders 57. The effects of sleep deprivation on clock-gene expression demonstrated here combined with the fact that sleep deprivation is a well known antidepressant add a molecular component to the intricate relationship between mood, circadian rhythms, and sleep (reviewed in 5859).

PF initiated and designed experiments, analyzed data, and wrote manuscript. RT conducted experiments, developed methods, and analyzed data. HCH was involved in experimental design, lab direction, and editing manuscript. BFO, along with PF, initiated and designed experiments, analyzed data, directed all experimental work by RT, and co-wrote manuscript. All authors have read and approved the final article.