Hypocretin (orexin) is critical in sustaining theta/gamma-rich waking behaviors that drive sleep need

Significance

As in narcolepsy patients, overall time spent awake and asleep is normal in mice lacking the wake-enhancing neuromodulator hypocretin/orexin. We discovered, however, that these mice, in baseline conditions, are impaired in maintaining theta-dominated wakefulness (TDW), a waking substate characteristic of goal-driven, explorative behaviors and associated with heightened θ/fast-γ activity. We demonstrate that TDW instability causes profound blunting of EEG δ activity in subsequent slow-wave sleep, a measure gauging homeostatic sleep need. In contrast, manually enforced waking induced unimpaired TDW expression and normal δ activity in recovery sleep. This suggests that TDW, not overall waking, drives sleep need, a hypothesis we verified by modeling the homeostatic process. We propose that Hcrt is critical for spontaneous waking but that enforced waking relies on other neuromodulators, such as norepinephrine.

Abstract

Hcrt gene inactivation in mice leads to behavioral state instability, abnormal transitions to paradoxical sleep, and cataplexy, hallmarks of narcolepsy. Sleep homeostasis is, however, considered unimpaired in patients and narcoleptic mice. We find that whereas Hcrt^ko/ko mice respond to 6-h sleep deprivation (SD) with a slow-wave sleep (SWS) EEG δ (1.0 to 4.0 Hz) power rebound like WT littermates, spontaneous waking fails to induce a δ power reflecting prior waking duration. This correlates with impaired θ (6.0 to 9.5 Hz) and fast-γ (55 to 80 Hz) activity in prior waking. We algorithmically identify a theta-dominated wakefulness (TDW) substate underlying motivated behaviors and typically preceding cataplexy in Hcrt^ko/ko mice. Hcrt^ko/ko mice fully implement TDW when waking is enforced, but spontaneous TDW episode duration is greatly reduced. A reformulation of the classic sleep homeostasis model, where homeostatic pressure rises exclusively in TDW rather than all waking, predicts δ power dynamics both in Hcrt^ko/ko and WT mouse baseline and recovery SWS. The low homeostatic impact of Hcrt^ko/ko mouse spontaneous waking correlates with decreased cortical expression of neuronal activity-related genes (notably Bdnf, Egr1/Zif268, and Per2). Thus, spontaneous TDW stability relies on Hcrt to sustain θ/fast-γ network activity and associated plasticity, whereas other arousal circuits sustain TDW during SD. We propose that TDW identifies a discrete global brain activity mode that is regulated by context-dependent neuromodulators and acts as a major driver of sleep homeostasis. Hcrt loss in Hcrt^ko/ko mice causes impaired TDW maintenance in baseline wake and blunted δ power in SWS, reproducing, respectively, narcolepsy excessive daytime sleepiness and poor sleep quality.

Wakefulness encompasses a wide spectrum of arousal levels, sensorimotor processing modes, and behaviors, reflected in the electroencephalogram (EEG) by a great variety of signal patterns (1). Fourier transform decomposes the signal into multiple frequency ranges, defining δ, θ (5 to 10 Hz), and γ (>30 Hz) oscillatory components. However, although the EEG has been used to define wakefulness and distinguish it from slow-wave and paradoxical sleep (SWS and PS) for decades, a formal cartography of waking substates, associated EEG features, and their significance for the animal is largely lacking. Likewise, definition of the relation between waking quality and subsequent sleep content has evolved little since enunciation of the two-process model of sleep regulation (2) in which a sleep/wake-dependent homeostatic “process S,” reflected in EEG δ power during SWS, regulates sleep propensity as a function of prior waking duration regardless of waking quality. Later studies described behaviors (3, 4) or EEG components (5, 6) that disproportionally affect the sleep homeostat. Different procedural, cognitive, or emotional experience differentially impacts subsequent SWS δ power not only quantitatively but also spatially across brain areas, hence heralding sleep as a local, use-dependent process (7). If wake partly instructs sleep, is a formal modeling of the link between waking spectral dynamics and sleep spectral dynamics possible? Such insight may greatly further our understanding of sleep-need physiology.

Waking relies on wake-active cell populations in a set of key arousal centers (8). Among them, Hcrt cells in lateral hypothalamus (9), noradrenergic cells in locus coeruleus (10), and dopaminergic cells in ventral tegmental area (11) can drive sleep-to-wake transitions. Hcrt cells send brain-wide projections that, in response to environmental cues, excite arousal nuclei by acting on one or both Hcrt receptors. Distinct wake-enhancing modulators promote distinct aspects of behavioral and cognitive arousal (11–13), but whether different wake-active cells or neuromodulator/receptor pairs differentially regulate waking electrocortical signatures or differently impact the sleep homeostat is poorly understood. In flies, striking differences are seen between wake induced by acetylcholine, dopamine, or octopamine cell activation on subsequent sleep (14). Among mammalian wake-enhancing neuromodulators, only Hcrt loss leads to a defined clinical condition, narcolepsy with cataplexy, characterized by sleep attacks, unstable nocturnal sleep, PS emergence into wake, and cataplexy. Hcrt knockout (KO) mice display all these symptoms, including an increase in state transitions, and thus model the behavioral state instability at the core of narcolepsy pathology (15). Untimed sleep, due to excessive daytime sleepiness, led early investigators to question the functional integrity of the sleep homeostat in narcolepsy. When patients were challenged by 24-h sleep deprivation (SD), subsequent time asleep and SWS δ power were, however, normal (16). Normal SWS δ power after 6-h SD was confirmed in the Hcrt^ko/ko mouse model (15).

We confirm this finding but discover that, in SWS following their spontaneously most active waking period, Hcrt^ko/ko mice display a severely reduced δ power, both relative to WT controls and to process-S simulation derived from prior sleep/wake history. This SWS δ power deficit is found to parallel reduced θ and fast-γ oscillatory dynamics in preceding wake. This led us to define a waking substate characterized by θ spectral dominance, theta-dominated wakefulness (TDW), which KO mice are unable to stably sustain in baseline and is associated with neural plasticity-related gene expression. In contrast, KO mice express normal TDW and θ/fast-γ activity when externally stimulated during SD. Using time in TDW to substitute time in all waking in process-S modeling, we predict SWS intensity solely based on TDW content of preceding wake. Thus, narcolepsy symptoms of daytime sleepiness and nighttime sleep fragmentation might be modeled in Hcrt^ko/ko mice by a spontaneously unstable TDW state in active phase and diminished magnitude of δ oscillations in ensuing SWS.

Results

Hcrt^ko/ko Mouse SWS Following Spontaneous Waking Shows Profound Blunting of EEG δ Power.

Consistent with prior reports (15), Hcrt^ko/ko mice responded to a 6-h SD with a prominent SWS δ power surge of similar dynamics and magnitude as WT mice. Surprisingly, however, in preceding baseline dark phases and night following SD, SWS δ activity of KO mice was severely diminished compared with WT controls (Fig. 1A, Top), although they both spent similar time in SWS (Fig. 1A, Bottom). Hcrt^ko/ko mouse SWS δ activity only modestly increases in the first 3 dark h, although this time interval witnesses maximal wakefulness (Fig. 3A) and locomotion (Fig. S1B) in KO as in WT mice.

Fig. 1.

Hcrt^ko/ko mice display a profound deficit in EEG δ (1.0 to 4.0 Hz) power in baseline dark-phase SWS but respond to SD with an SWS δ power rebound indistinguishable from WT littermates. (A, Top) Dynamics of SWS δ power in Hcrt^ko/ko and Hcrt^+/+ mice in the course of a 3-d recording with 6-h SD initiated at light onset of day 3 (ZT0 to 6). Mean δ power values (±SEM; KO, black symbols, n = 8; WT, gray symbols, n = 7) are expressed as the percentage of each mouse’s mean SWS δ power in baseline ZT8 to 12. Solid curves (KO, black; WT, gray) represent process-S simulations calculated using published equation parameters (17). (A, Bottom) SWS (min/h) time course across the 3 d. (B) EEG δ power dynamics at state transitions to SWS, in the minute preceding and the 2 min following, SWS onset (time 0) in four baseline intervals (Left) and four recovery intervals (Right). Similar δ power levels are quickly reached in both genotypes at all times, except in early dark (ZT12 to 18), when KO mice show a reduced δ power plateau. Orange bars, intervals with significant genotype differences in δ power (t test, P < 0.05). In B, statistics concern only time after “time 0.”

Fig. 3.

TDW expression is impaired in Hcrt^ko/ko mice. KO mice spend markedly less time in TDW in baseline dark phase than WT mice (A), due mainly to reduced TDW bout duration (B). They are, however, able to generate a TDW state spectrally similar to WT controls (C, Left), albeit of lower θ peak frequency (D) but higher slow-γ power (F). (A) Time courses of wake (min/h), TDW (min/h), and TDW as the percentage of total wake in Hcrt^ko/ko (black; n = 8) and Hcrt^+/+ (gray; n = 7) mice. (B) Distribution of TDW bout duration (Top), and time spent in different bout-duration categories (Bottom). (C) EEG spectra in TDW (Left) and all other waking (Right) in baseline. Power density values are normalized as in Fig. 2A. Differences at ∼10 Hz stem from the slower TDW θ peak frequency in KO mice relative to WT. (D) Blowup of TDW spectra of C, with focus on θ and slow-γ (Gamma-1) power. Asterisks indicate significant genotype difference in frequency showing maximal power density (t test, P < 0.05). (E) EEG spectra of three types of KO mouse waking: minute preceding cataplexy (pink), baseline dark TDW (ZT12 to 18; black), and enforced waking (6-h SD; blue). A slow-γ peak (∼40 Hz) is seen in precataplexy and baseline dark TDW but not in SD. Horizontal bars, frequency bins with significant power density differences between baseline TDW and SD (black), or precataplexy and SD (pink) (t test, P < 0.05). (F) To emphasize differences between baseline dark TDW and SD wake, the TDW/SD EEG power density ratio is plotted across EEG frequency. Orange bars and asterisks indicate measures showing significant genotype differences (t test, P < 0.05).

Fig. S1.

TDW can uncouple from locomotor activity. (A) Time-course analysis of a representative WT mouse (OX15) for LMA (pink; movement counts per min; range 0 to 35), TDW (black; percentage of 4-s epochs per min; range 0 to 100%), and all waking (gray; percentage of epochs per min) across the 3-d recording. (A, Top and Middle) Baseline days 1 and 2. (A, Bottom) SD and recovery (REC). (B) TDW and LMA hourly mean (±SEM) values for Hcrt^ko/ko (KO; black) and Hcrt^+/+ (WT; gray) mice (TDW, n = 7 and 8 for WT and KO, respectively; LMA was recorded in a subset, n = 4 and 5 for WT and KO, respectively). Orange bars, time intervals with significant genotype differences (t test, P < 0.05). (C) The LMA–TDW relation in B is quantified by linear regression. Dots represent hourly values for mice in which TDW and LMA were simultaneously acquired (WT, gray, n = 4; KO, black, n = 5; 72 values per mouse). Pairs of regression lines (gray and black for WT and KO, respectively) delimit the 95% confidence interval (CI) of the correlation within each genotype. Although the correlation is strong and highly significant in both (P < 0.0001; R² = 0.73 and 0.87, for KO and WT, respectively), the regression line slope is significantly steeper in WT mice [8.7 ± 0.2 vs. 5.8 ± 0.2 (movements per min); lack of overlap of 95% CI]. (D, Top) TDW prevalence during LMA is assessed by determining the occurrence of ≥1 TDW 4-s epoch within minutes in which LMA count is >0 (expressed in %). (D, Bottom) LMA prevalence during TDW by determining the occurrence of ≥1 LMA count within minutes in which ≥1 TDW epoch is observed. Separate analyses in baseline light and dark phases revealed that when LMA is present, chances are high (57 to 75%) that TDW is expressed. This percentage did not differ between genotypes and lighting condition (two-way ANOVA; factor genotype, P = 0.25; LD, P = 0.91; interaction, P = 0.69). TDW occurrence was, however, less often associated with LMA in the light compared with the dark, and in KO compared with WT mice (two-way ANOVA; factor genotype, P = 0.0042; LD, P < 0.0001; interaction, P = 0.63). Hence, TDW appears as a more sensitive marker of active waking than is LMA. Orange asterisks indicate genotype differences (t test, P < 0.05); the black asterisk indicates light vs. dark differences (paired t test, P < 0.05).

What could explain KO mouse profound deficit in SWS δ activity after being awake in some contexts but not others? To assess whether this reflects KO mouse slightly lower total waking time in the first half of the night compared with WT mice (Fig. 3A), thus predicting lower sleep need, we applied a mathematical simulation of the sleep homeostatic process S (2, 5). These initial simulations were run using parameters previously determined in mice of the same genetic background (17) (Methods). Simulations reproduced remarkably well the actual δ power values in WT mice (Fig. 1A, Top). In KO mice, however, actual values fell well below the simulation in dark phases. Hcrt loss thus severely disrupts the relationship between time spent spontaneously awake and subsequent SWS δ power.

We next reasoned that the shorter duration of SWS bouts in KO mice (Fig. S2B) may preclude a high-power δ rhythm to set in. We thus analyzed δ power kinetics surrounding SWS onset. In both genotypes, high-amplitude δ power was seen to rapidly set in, reaching a plateau in less than 30 s (Fig. 1B), thus ruling out SWS fragmentation as a cause for KO mouse δ power deficit. Because KO mouse baseline wake fails to raise sleep drive normally, we next set to determine whether alteration of its quality causes its different homeostatic weight.

Fig. S2.

Time course of time spent asleep and distribution of sleep-bout duration in Hcrt^ko/ko and Hcrt^+/+ littermate mice. (A) Time course of SWS (min/h) (Top) and PS (min/h) (Bottom) in the 3-d recording, including baseline (0 to 48 h), a 6-h SD initiated at light onset of the third day (ZT0 to 6), and recovery (54 to 72 h). Mean values (±SEM; KO, black filled circles, n = 7; WT, gray filled circles, n = 8). SWS data as in Fig.1A. Hcrt^ko/ko mice spent more time in PS than Hcrt^+/+ mice, in particular in the dark-phase first half. Orange bars, significant genotype differences (t test, P < 0.05). (B) Analysis of distributions of SWS (Top) and PS (Bottom) episode duration (Left) and time spent in different categories of bout duration (Right) reveals a higher prevalence of shorter episodes in both SWS and PS in mutant mice. Asterisks indicate bout lengths with significant genotype differences (t test, P < 0.05).

Blunting of θ and Fast-γ Oscillations in Dark-Phase Spontaneous Wake of Hcrt^ko/ko Mice.

Dynamic analysis of the full waking EEG spectrum in the 3-d recording reveals that most salient changes occur in θ (6.0 to 9.5 Hz), slow-γ (32 to 45 Hz), and fast-γ (55 to 80 Hz) (18) (Fig. 2B). Behaviorally most active periods, namely early night and SD, display a sharp concerted increase in θ and fast-γ power and a modest slow-γ increase in both genotypes (Fig. 2C). Hcrt^ko/ko mice, however, display greatly diminished θ (Fig. 2C, Top) and fast-γ (Fig. 2C, Bottom) power in baseline dark relative to WT, whereas slow-γ is unchanged (Fig. 2C, Middle). During SD, in contrast, which features the highest θ/fast-γ power of the 3 d, KO mice display θ/fast-γ power surges of similar magnitude as WT mice (Fig. 2C).

Fig. 2.

Waking EEG θ (6.0 to 9.5 Hz) and fast-γ (55 to 80 Hz) oscillatory activities of Hcrt^ko/ko mice are markedly reduced in baseline dark phase but similar to WT mouse values in most of the 6-h SD. (A) EEG spectra of Hcrt^ko/ko (n = 8) and Hcrt^+/+ littermates (n = 7) in (Top) dark-phase spontaneous waking (ZT12 to 18) and (Bottom) enforced waking (ZT0 to 6). EEG power density values are expressed as the percentage of each mouse’s total baseline EEG power (SI Methods). Note the nonlinear axes. (A, Top, Inset) Blowup of EEG spectra in the 0- to 16-Hz range. Axes here are linear. Orange bars, frequency bins with significant genotype differences in power density (t test, P < 0.05). (B) Heatmap of the spectral dynamics of the waking EEG across the 3 recorded days. Color-coded are mean waking EEG power density values across frequency bins, relative to each mouse’s mean power density for that bin in waking of the baseline light-phase last 4 h (ZT8 to 12). Contour lines connect (frequency;time) pairs of equal relative EEG power density at 30% increments. Activity around 50 Hz (47.5 to 52.5 Hz) was eliminated and linearly interpolated. Changes are most pronounced in θ, slow-γ, and fast-γ power. Changes at ∼10 Hz result from an abrupt increase in TDW θ peak frequency at dark onset (Fig. S5). (C) Waking θ, slow-γ (Gamma-1), and fast-γ (Gamma-2) EEG power time course. Power value (±SEM) in the respective frequency range is expressed as in B. Orange bars, periods with significant genotype differences (t tests, P < 0.05).

We next compared waking spectra in baseline early dark [Zeitgeber (ZT)12 to 18], when KO mice show severe SWS δ deficit, and 6-h SD, followed by normal SWS δ power. In baseline dark, WT mouse waking shows an ∼8-Hz θ peak that is absent in KO, in which an ∼5-Hz peak prevails (Fig. 2A, Top). During SD, in contrast, spectra feature a prominent ∼8-Hz θ rhythm in both genotypes (Fig. 2A, Bottom).

A Theta-Dominated Waking Substate.

Rodent waking is often divided into quiet and active wake based on whether the animal is still or exhibits exploratory movements (1, 19). Transitions from quiet to active wake correlate with increased θ and γ power. Moreover, correlations between locomotion, waking θ, and subsequent SWS δ power are reported (4, 6), three variables found reduced in spontaneous waking of KO mice (for locomotion, see Fig. S1), suggesting that Hcrt has a role in active waking.

The blunting of θ/fast-γ power seen in baseline KO mouse waking might be due to spending less time in θ/fast-γ–rich wake or to a global weakened ability to generate θ/fast-γ oscillations. To examine this, we designed an EEG-based algorithm that quantifies active waking, identified by the regular stereotypical θ rhythmic pattern, as seen in rodents engaged in goal-driven exploration (Fig. S3 and Methods). Video recordings were used to fine-tune defining parameters. The state thus identified was called theta-dominated wakefulness following a previously coined term (20).

Fig. S3.

Defining the TDW state: impact of the θ/total EEG power ratio threshold chosen to define TDW on time spent in TDW, θ peak frequency, and θ peak power. (A) The θ/total EEG power ratio was calculated for all waking 4-s epochs in which maximal EEG power falls in the θ range (see Methods for details). The distribution of θ/total power ratio in these “candidate-TDW” epochs is depicted as the percentage of the sum of all of them (100%; i.e., the area under the curve). The distributions in the two genotypes closely overlap. Visual inspection of the EEG trace pattern recorded in a subset of mice revealed that with a θ/total power threshold of 0.228, the algorithm adequately identifies waking epochs in which EEG is dominated by θ activity. This threshold rejects 12.3% of all candidate TDW epochs. For comparison, distribution of the θ/total power ratio in PS is depicted. The ratio is clearly larger in PS than in TDW, reflecting the visually “cleaner” θ rhythm in PS. Again, the distribution does not differ between genotypes. Applying the 0.228 threshold to PS would have rejected 1.6% of all epochs visually scored as PS. (B) The pronounced deficit in time spent in TDW of Hcrt^ko/ko mice (KO, black lines) relative to Hcrt^+/+ littermates (WT, dark gray lines) persists over a wide range of θ/total power ratio thresholds (lighter gray areas ± SEM). (C) Similarly, the slower TDW θ peak frequency of Hcrt^ko/ko mice is independent of the threshold used. Interestingly, TDW epochs with a higher θ/total power ratio show a faster θ peak frequency. (D) The similarity in TDW θ peak power between the two genotypes does not depend on this threshold either. Orange horizontal bars, significant genotype differences (t test, P < 0.05).

Baseline TDW spectra of Hcrt^ko/ko and Hcrt^+/+ mice were remarkably similar (Fig. 3C, Left). Furthermore, when TDW’s θ and γ components were dynamically examined across the 3 d (as done above for all waking in Fig. 2C), we found the severe genotype differences to mostly vanish (Fig. S4). Namely, (i) the steep fluctuations in waking θ and fast-γ power across time and conditions were much reduced within the TDW state, and (ii) the pronounced genotype difference disappeared in θ and markedly diminished in fast-γ (Fig. S4). This strongly supports the efficacy of the TDW algorithm to capture a homogeneous waking substate, which is stable both among the two genotypes and across time and behavioral conditions.

Fig. S4.

TDW state’s EEG power dynamics in the θ (6.0 to 9.5 Hz), slow-γ (32 to 45 Hz), and fast-γ (55 to 80 Hz) ranges. Mean power values (±SEM) in the respective frequency ranges are expressed as the percentage of each animal’s mean θ, slow-γ, or fast-γ power in TDW of the last 4 h of baseline light phase (ZT8 to 12) in Hcrt^ko/ko (KO, black symbols) and Hcrt^+/+ (WT, gray symbols) mice. Orange bars, time intervals with significant genotype differences (t test, P < 0.05). To facilitate comparison with the same analysis performed for wakefulness instead of TDW, axes use the same scaling as in Fig. 2B. Note that the pronounced changes over time and genotypes observed in Fig. 2B for θ and fast-γ activity are no longer present when analyzed specifically in the TDW substate. This analysis demonstrates that most differences in waking EEG stem from time and genotype differences in waking TDW state content, not from intrinsic differences in generating these oscillations. This confirms that Hcrt^ko/ko mice are able to generate a by and large normal TDW state but are impaired in stabilizing this activity over time.

Even though KO mouse TDW is qualitatively by and large normal, detailed analysis reveals differences. Whereas TDW in the two genotypes displays similar θ power (KO 1.74 ± 0.12% vs. WT 1.86 ± 0.18%; P = 0.57, t test), θ peak frequency is slower by half a Hz in KO mice (Fig. 3D, Left; KO 7.72 ± 0.10 vs. WT 8.21 ± 0.11; P = 0.006, t test). This KO mouse θ peak shift causes the ∼10-Hz power density deficit seen in Fig. 3C (Left) and also in the SD waking EEG (Fig. 2A, Bottom).

Applying the TDW algorithm across the 3 d revealed that, in their spontaneously most active phase (baseline dark), Hcrt^ko/ko mice spend markedly less time (min/h) and fraction of total wake (%) in TDW, compared with WT controls (Fig. 3A). WT mice spent in total 3.55 ± 0.41 h per 12-h baseline night phase in TDW (30%), whereas KO mice spent only 2.05 ± 0.13 h in TDW (17%) (P = 0.003, t test). In contrast, during the 6-h SD, time in TDW did not differ (3.18 ± 0.33 and 2.98 ± 0.32 h, in WT and KO, respectively; P = 0.67, t test; Fig. 3A). KO mouse diminished baseline TDW was due to both a reduced rate of W-to-TDW transitions (−20%) and impaired TDW bout duration (−20%; Fig. 3B). A great number of TDW episodes lasted only for one epoch (4 s) (52 and 44%, in KO and WT mice, respectively; Fig. 3B) and, although the number of 4-s TDW bouts was similar in the two genotypes (KO 617; WT 651; Table S1 and Fig. 3B, Top), TDW bouts lasting longer than 4 s were over 40% fewer in KO mice than in controls (P = 0.0006, t test). Hcrt^ko/ko mice thus appear impaired in effecting stable wake-to-TDW transitions and maintaining TDW network activity beyond 4 s.

Table S1.

TDW episode number and duration (mean ± SEM) in baseline conditions (48-h) in Hcrt^ko/ko (n = 8) and Hcrt^+/+ (n = 7) mouse littermates

	All TDW episodes		TDW episodes > 4 s		TDW episodes = 4 s	No. of 4-s episodes per all TDW episodes,
Mice	No. of episodes per 24 h	Episode duration, s	No. of episodes per 24 h	Episode duration, s	No. of episodes per 24 h	%
Hcrtko/ko	1,193 ± 49	8.3 ± 0.3	576 ± 27	13.0 ± 0.4	617 ± 23	51.8 ± 0.6
Hcrt+/+	1,487 ± 70	10.4 ± 0.7	836 ± 54	15.3 ± 1.1	651 ± 19	44.0 ± 1.1
P	0.0038	0.014	0.0006	0.053	0.29	<0.0001

Hcrt^+/+ mice express 25% more TDW episodes, and TDW episodes are on average 25% longer relative to Hcrt^ko/ko littermates. When considering only TDW episodes longer than 4 s, this genotype difference in bout number greatly increases (WT express 45% more episodes), whereas episode duration no longer significantly differs. The number of single 4-s TDW bouts is astonishingly similar in the two genotypes (see also Fig. 3B, Top). Thus, in Hcrt^ko/ko mice, the percentage of TDW episodes that last only 4 s is greatly increased. We conclude that Hcrt^ko/ko mice are greatly impaired in stably maintaining TDW beyond 4 s. The last row indicates P values for genotype differences (t test).

In sum, Hcrt loss largely preserves the integrity of the TDW state EEG signature but compromises its stability, resulting in overall TDW state paucity. Because (i) TDW spectra are by and large normal, (ii) time spent in spontaneous TDW is reduced in KO mice, and (iii) genotype differences in θ/fast-γ dynamics disappear when analyzed within the TDW EEG, we conclude that the pronounced genotype differences in waking θ/fast-γ dynamics are not due to a reduced capacity to generate these rhythms but to engaging in a substate that supports their expression.

A Behavioral Validation of TDW.

As mentioned above, Hcrt^ko/ko mice reproduce the pathognomonic narcolepsy symptom of cataplexy. Cataplexy-triggering behaviors in mice are those typically associated with TDW, namely shelter construction and exploration. A behavioral analysis of the recordings presented here used EEG-independent, video-assessed criteria to identify cataplexy (21). Epochs that precede cataplexy typically match our video-independent, EEG-based TDW scoring criteria, thus offering a behavioral validation of the TDW definition. To further validate this relationship, we compared Hcrt^ko/ko mouse EEG spectra in (i) baseline dark TDW, (ii) the minute preceding cataplexy, and (iii) 6-h SD waking (Fig. 3E). The three spectra almost match each other, all featuring a major ∼8-Hz power peak. A notable difference is that baseline dark TDW and precataplexy both share an additional peak at ∼40 Hz (slow-γ) that is absent in enforced wake. Hcrt^+/+ mice are found to also express a slow-γ peak at similar frequency in baseline dark TDW (Fig. 3 C and D), although of a significantly lower power than KO (Fig. 3F). This may be relevant to narcolepsy symptomatology, as β and slow-γ synchrony is seen in dopamine-depleted mice performing specific tasks and can correlate with motor dysfunction (22). We thus demonstrate that baseline dark and precataplexy define two spontaneous waking paradigms that rely on Hcrt to (i) stably express TDW, (ii) maintain normal TDW θ oscillatory frequency, and (iii) limit a concomitant slow-γ component.

TDW Is the Principal Driver of the Sleep Homeostat.

Because Hcrt^ko/ko mouse wake induces normal SWS δ power when its TDW content is normal (SD) but not when its TDW content is abnormally low (baseline), we sought to formally determine whether time in TDW more accurately predicts SWS intensity than total waking. To this end, we simulated process-S kinetics assuming S increases only in TDW rather than in all waking.

In initial simulations (Fig. 1A), we used for both genotypes parameters defined previously (17) (Table 1, model 1). Here we further optimized the parameters in each mouse to best fit the actual δ power measures. Assuming S increases in wake and PS, S evolves as in Fig. 4B (Top, solid curves) with equation parameters as in Table 1 (model 2). These S values match actual values with an excellent fit at all time points in WT mice. In KO mice, however, the simulation fails, in particular in light phase, as it “attempts” to fit the low δ power values measured in dark phase by underestimating expectation in light phase. Moreover, values obtained for τi and τd significantly differ in KO and WT mice. To assume that S increases in all types of wake (and PS) thus leads to invoking an alteration of processes regulating sleep homeostasis in KO mice. In contrast, modeling assuming that S increases solely in TDW predicts S increase and decrease rates that are similar in both genotypes (Table 1, model 3) and a better fit in KO mice (Fig. 4B, Bottom), whereas the fit stays excellent in WT mice. Because KO mice respond to 6-h SD with an SWS δ power rebound similar to WT controls, and because no evidence exists for altered process-S kinetics in narcolepsy (15, 16), an alteration of τi and τd rates in KO mice, as would result from model 2, is not favored. Our data most fit the hypothesis that S increases specifically in TDW and that sleep homeostatic processes are unimpaired in KO mice (model 3). Sleep homeostatic drive is thus primarily caused by events occurring in TDW, which accounts for only 18% of baseline recording [WT 4.38 ± 0.52 h (18.2%); KO 2.78 ± 0.19 h (11.6%); P = 0.0094, t test].

Table 1.

Equation parameters used in process-S modeling

Parameters	KO	WT	P
Model 1: Process S increases when not in SWS (in W+PS)
Parameters as in Franken et al. (17) (Fig. 1A, Top)
τI, h	7.9	7.9	—
τd, h	1.9	1.9	—
LA, %	55	55	—
UA, %	282	282	—
Fit, Δ%	23.9 ± 1.9	17.3 ± 2.5	0.049*
Model 2: Process S increases when not in SWS (in W+PS)
Parameters individually optimized (Fig. 4B, Top)
τI, h	17.6 ± 1.0	10.3 ± 1.7	0.0025*
τd, h	2.4 ± 0.2	1.7 ± 0.2	0.0094*
LA, %	47 ± 8	61 ± 6	0.20
UA, %	418 ± 22	361 ± 36	0.19
Fit, Δ%	14.0 ± 2.0	9.2 ± 1.3	0.078
Model 3: Process S increases in TDW exclusively
Parameters individually optimized (Fig. 4B, Bottom)
τI, h	4.3 ± 0.4	3.5 ± 0.4	0.13
τd, h	1.8 ± 0.3	1.4 ± 0.1	0.34
LA, %	80 ± 6	89 ± 3	0.24
UA, %	331 ± 22	310 ± 26	0.55
Fit, Δ%	12.7 ± 1.8	10.1 ± 1.7	0.33

Parameters yielding the best fit with actual EEG δ power dynamics are listed (mean ± SEM; KO, n = 8; WT, n = 7). In model 1 (Fig. 1A, Top), S is estimated using published parameters (17). For models 2 (Fig. 4B, Top) and 3 (Fig. 4B, Bottom), parameters were individually optimized in each mouse. Whereas S increases in W and PS in models 1 and 2, S increases exclusively in TDW in model 3. The third column lists P values for differences in parameter values between the two genotypes (underscored; t test, *P < 0.05). To quantify fit (Δ%), the mean % difference between simulated and observed δ power values across all time points is listed. Only model 3 yields a close fit in both genotypes and with similar parameters.

Fig. 4.

Process-S modeling. (A) S is calculated iteratively in each 4-s epoch, based on the mouse’s behavioral state sequence and two equations that feature four parameters. The S buildup equation is applied in W and PS (B, Top; model 2; Table 1) or exclusively in TDW (B, Bottom; model 3), and features an S increase rate (τi) time constant and an upper asymptote (UA; dashed line). The S decay equation is applied only in SWS in both models, and features an S decrease rate (τd) time constant and a lower asymptote (LA; dashed line). In each mouse, the four-parameter set yielding S values best fitting actual EEG data is determined (Table 1 for parameters yielding the best fit). (B) Time course of process-S simulation and actual EEG δ power across the 3 d. Round symbols (black, KO, n = 8; gray, WT, n = 7) depict mean (±SEM) relative EEG δ power values. Curves depict mean process-S values. The “classical” model (Top) fails for Hcrt^ko/ko mice in baseline light phase, and parameter best fitting asks for S buildup and decay rates to be slower in KO than in WT mice (Table 1). Fit markedly improves, and the S buildup and decay rate best-fitting parameters no longer differ between genotypes, when S is assumed to increase solely in TDW (Bottom). Blue bars, significant differences between measured and simulated values in Hcrt^ko/ko mice (t test, P < 0.05). No differences are seen in Hcrt^+/+ mice.

Hcrt^ko/ko Mouse δ Deficit in SWS Affects Predominantly the Slow-δ Frequency Range.

We found that the SWS δ power deficit of KO mice in baseline dark is not equally distributed across the δ (1.0 to 4.0 Hz) range but mostly impacts slow-δ (δ1; 1.0 to 2.25 Hz) whereas fast-δ (δ2; 2.5 to 4.0 Hz) is less impaired, resulting in a greatly increased fast/slow (δ2/δ1) power ratio relative to WT (Fig. 5 A and B, Left). In contrast, the first 20 min of SWS after SD display high power in both δ components in KO as in WT mice, thereby normalizing KO mouse δ2/δ1 ratio (Fig. 5 A and B, Right). Interestingly, an SWS slow-δ deficit is also seen in recovery sleep of norepinephrine-depleted rats, and found, as shown below in our mice, to correlate with reduced expression of neural plasticity genes in preceding wake (23).

Fig. 5.

SWS δ power deficit of Hcrt^ko/ko mice in baseline dark phase affects predominantly slow-δ oscillations. (A) SWS EEG spectra in the first 6 h of baseline dark phase (ZT12 to 18; Left) and in the first 20 min of recovery after SD (Right) in Hcrt^+/+ (WT; gray, n = 7) and Hcrt^ko/ko (KO; black, n = 8) mice. In baseline dark SWS, genotype differences are more pronounced in slow (δ1; 1.0 to 2.25 Hz) than in fast-δ (δ2; 2.5 to 4.0 Hz) power, whereas in recovery, WT and KO spectra largely overlap. Spectra are normalized as in Fig. 2A. Note the nonlinear frequency axis. (B, Left) In baseline SWS, KO mice show reduced power in both δ bands relative to WT but, unlike WT mice, power in slow-δ is significantly lower than in fast-δ (left ordinate; two-way rANOVA; genotype, P = 0.0046; δ band, P = 0.0002; interaction, P = 0.049), resulting in a vastly larger fast-to-slow δ power ratio (δ2/δ1 ratio; right ordinate) in KO mice. (B, Right) In recovery SWS, both fast- and slow-δ power, as well as fast-to-slow ratio, are restored to WT levels (two-way rANOVA; genotype, P = 0.51; δ band, P = 0.024; interaction, P = 0.56). Orange asterisks, power differences between genotypes (t test, P < 0.05); black asterisk, fast- vs. slow-δ power difference within genotype (paired t test, P < 0.05).

TDW: Not Just a Proxy of Locomotor Activity!

Because prominent θ typically accompanies locomotion, we analyzed the interdependency of the TDW state and locomotor activity (LMA). As previously reported (24), locomotion in baseline dark phase is reduced in Hcrt^ko/ko mice (Fig. S1B). During SD, LMA rose to higher counts than in early dark phase in both genotypes, but KO mice remained less active than WT controls. Linear regression indicated that TDW and LMA are globally strongly correlated in both genotypes, although the slope of the relation is significantly less steep in KO mice (Fig. S1C), indicating that not only do these mice spend less time in TDW but they are also less active per unit of time in TDW. Examination of the TDW–LMA relationship across the 3 d, however, revealed discrepancies. Whereas LMA is reduced in KO mice both in spontaneous and enforced waking (Fig. S1B), TDW expression is impaired in baseline but fully normal during SD. Thus, TDW can uncouple from movement. A high-resolution comparison of TDW and LMA in a representative WT mouse further supports this conclusion (Fig. S1A). We next assessed TDW prevalence during waking with LMA, and LMA prevalence in the TDW state (Fig. S1D for details). Results showed that TDW is a better predictor of LMA (Fig. S1D, Bottom) than LMA is a predictor of the TDW state (Fig. S1D, Top), as TDW monitoring could capture both the genotype and dark/light differences in LMA expression, whereas LMA monitoring could not. Altogether, we conclude that time in TDW, not LMA, correlates with SWS EEG δ power, and thus is potentially causally linked to sleep homeostatic drive.

Hcrt^ko/ko Mice Show Reduced Cortical Expression of Neuronal Activity-Related Genes in Early Dark Phase.

We next examined whether reduced TDW expression and sleep homeostatic drive correlate with altered cortical expression of neuronal activity-induced genes or genes whose expression increases with explorative behavior (4) or enforced waking (25). Arc, Bdnf, Egr1 (also known as NgfiA or Zif268), Fos, Homer1a, Hspa5 (also known as Bip), and Per2 transcripts were assayed either 1 h before dark onset (ZT11), when mice are fully rested after the major sleep period and SWS δ power is lowest, or 3 h into the dark phase (ZT15), when SWS δ power shows maximal deficit in Hcrt^ko/ko mice. To further understand the impact of external stimulation on waking variables, mice were either subjected to a 3-h “SD” from dark onset to ZT15 (group ZT15+SD) or left undisturbed (group ZT15−SD).

In late resting phase (ZT11), levels of all seven transcripts were low, and indistinguishable in KO and WT mice (Fig. 6). Up-regulation from ZT11 to ZT15 was observed for all transcripts in WT mice, as expected, as ZT 12 to 15 h witnesses maximal waking time in WT and KO mice (Fig. 3A). In KO mice, however, Bdnf transcript failed to increase, whereas the other six transcripts did show up-regulation, but levels reached at ZT15 were moderately, albeit as a group highly significantly, lower than WT mouse levels. In single-transcript comparisons, Bdnf, Egr1, and Per2 mRNAs showed significantly lower ZT15 relative abundance in KO than WT mice (Fig. 6).

Fig. 6.

Cortical expression of seven genes known to be up-regulated after prolonged waking. mRNA level was quantified in three groups of Hcrt^ko/ko (black) and Hcrt^+/+ (gray) mice (n = 7 to 9 per group per genotype). Two groups were left undisturbed and killed either 1 h before (ZT11) or 3 h after dark onset (ZT15−SD). A third group (ZT15+SD) was sleep-deprived between ZT12 and 15 and then killed. Values (±SEM) are expressed as the percentage of mean mRNA level at ZT11. Gene expression increased significantly from ZT11 to 15 in both genotypes, except if labeled ns (nonsignificant). Transcript levels in ZT15−SD were lower in KO than in WT mice (two-way rANOVA; factor genotype, P = 0.031; factor transcript, P < 0.0001; interaction, P = 0.29), although post hoc test is significant for Egr1, Per2, and Bdnf only (orange asterisks; t test, P < 0.05). SD increased levels further (group ZT15+SD vs. ZT15−SD), and this effect was more pronounced in KO than in WT mice (two-way ANOVA; factor group, P < 0.005 for all transcripts, except Egr1, P = 0.94; factor genotype, P > 0.29 for all transcripts, except Per2, P = 0.017; interaction: Egr1, P = 0.0022; Bdnf, P = 0.016; Arc, Fos, and Hspa5, P < 0.11) (9). ZT15+SD mRNA levels did not differ between genotypes, indicating that SD normalizes KO mouse phenotype. (Bottom Right) Process S is modeled, assuming S increases only in TDW. For ZT15+SD data points, the simulation uses EEG data collected across ZT12 to 15 in C57BL/6J WT mice sleep-deprived across ZT12 to 18 (77), assuming that TDW expression during SD does not differ between genotypes, as for ZT0 to 6 SD (Fig. 3A).

The 3-h SD condition up-regulated all transcripts in KO mice (group ZT15+SD vs. ZT15−SD), except Egr1, which in both WT and KO mice remained at ZT15 level irrespective of SD. It is noteworthy that external stimulation has the effect of normalizing the deficit of KO mice in baseline dark, not only at the EEG level (TDW expression, SWS δ power, δ2/δ1 ratio) but also at the molecular level, as all seven transcripts reached WT levels in KO mice after 3-h SD.

We then simulated process S for these experimental conditions, assuming S builds up in TDW only (Fig. 4B, Bottom and Table1, model 3). Results revealed how the dynamics of the seven transcripts tend to parallel those of the simulated sleep homeostatic drive (Fig. 6), with Bdnf most closely matching process S, consistent with prior reports (4).

Discussion

Here we provide an EEG-based algorithm that allows quantifying an active waking state, TDW, that we show by modeling to drive wake-dependent increase in sleep need and SWS δ power. We moreover show that, in spontaneously motivated arousal, TDW stability critically relies on Hcrt, possibly through Hcrt-dependent neuromodulation of circuits generating θ and fast-γ oscillations. Hence, Hcrt loss in Hcrt^ko/ko mice results in TDW state instability and reduced SWS intensity that may model, respectively, the excessive daytime sleepiness and poor sleep quality of narcolepsy.

Hcrt Is an Essential Regulator of the EEG Determinants of Spontaneous Waking.

We uncovered previously unseen alterations in spontaneous waking of Hcrt^ko/ko mice by discovering unexpected features in their sleep EEG. We found baseline SWS δ power to be severely blunted relative to WT littermates and predictions of sleep’s classic homeostasis model, in which SWS δ power (often conceptualized as “sleep need”) increases with time awake and decreases with time in SWS (2, 5). Hcrt^ko/ko mice were known to normally respond to an SD challenge (15); however, they were not previously thoroughly tested in spontaneous conditions.

Discrepancy between actual and process-S modeling-derived δ power suggested that Hcrt loss alters either (i) sleep homeostasis or (ii) spontaneous waking quality. Because Hcrt^ko/ko mice mount a fully normal SWS rebound after SD, the latter explanation was favored. Moreover, SWS δ power was shown to depend not only on prior time awake but also on EEG activity and behavioral content (4, 6, 26). Stress incurred by social defeat, for instance, enhances SWS δ power (3). Furthermore, studies on the genetic determinants of sleep homeostasis revealed that SWS δ power buildup rate is positively correlated with mean waking EEG θ power in mouse inbred lines (5). We thus examined the spectral content of the waking EEG in KO and control mice, revealing severe blunting of θ and fast-γ power in baseline dark phase in KO mice. Both rhythms, however, reach maximal power during SD with similar relative intensity in KO as WT mice, indicating that mechanisms generating θ and fast-γ oscillations are functional and recruited through alternate pathways in Hcrt absence. Spontaneous wake thus drastically differs from “gentle handling”-enforced wake in Hcrt dependence, presumed arousal circuits, and homeostatic impact. This is reminiscent of studies in Drosophila suggesting that distinct arousal neuromodulators impart distinct processing modes in target circuits, with distinct homeostatic costs (14). Likewise, Hcrt involvement in arousal appears to be driven by context-dependent determinants and drive system-wide processes, such as θ and fast-γ network rhythms.

Impaired TDW Stability and Cataplexy in Hcrt^ko/ko Mice.

The blunted EEG θ dynamics displayed by KO mice while spontaneously awake suggested that active waking, known to display a dominant θ rhythm, is specifically impaired in Hcrt absence. To verify this hypothesis, we designed the TDW scoring algorithm, which revealed that KO mice spend spontaneously markedly less time in TDW than WT mice, whereas time in TDW is normal during SD. Reduced baseline TDW is due to reduced TDW bout number and duration, indicating that Hcrt loss impairs the ability to maintain the TDW state for 4 s or longer.

Hcrt^ko/ko mice display many features of human narcolepsy, including cataplexy. It should be noted that TDW fragmentation in these mice is not a consequence of cataplexy attacks interrupting TDW bouts. Just as these mice generate bouts of spectrally normal TDW in early dark phase, they demonstrate normal TDW-associated behaviors (vigorous running, burying, nest building), which are cataplexy triggers. Cataplexy thus interrupts a fraction of KO mouse TDW bouts, and we indeed showed precataplexy waking to be enriched in EEG θ and γ activity relative to global dark-phase waking (21). However, overt cataplexy remains rare (ca. 23 episodes per 12-h dark phase) (21), compared with TDW episodes (1,193 per 24 h). Although we cannot formally rule out that TDW fragmentation is linked to motor dysfunction below the sensitivity of our cataplexy definition, state instability is a hallmark phenotype of Hcrt^ko/ko mice (15), and evidence favors TDW fragmentation to result from intrinsic TDW state lability in Hcrt absence.

Updating Sleep-Need Modeling.

Existence within the same animal of two types of wake with distinct sleep homeostatic weights provides an opportunity to probe for attributes of wakefulness, including EEG variables, substate composition, and underlying neural determinants, that link wake to the sleep homeostat and thus determine sleep need and recovery function. To formally address whether KO mouse TDW deficit accounts for their SWS δ deficit, we turned to process-S equations (5), substituting wake for TDW, that is, applying the S buildup term only in TDW and S decay term only in SWS. This yielded (i) similar values in KO and WT mice for S buildup and decay rates and upper and lower S asymptotes, consistent with the notion of intact sleep homeostatic processes in KO mice, and (ii) faithful SWS δ power value prediction for both genotypes and both waking paradigms.

Although process-S modeling fits WT mouse data almost perfectly, KO mouse values remain less fitted, in particular in light phase. How can our modeling be further improved? First, we show that KO mouse TDW has a slower θ rhythm and γ-range differences relative to WT mice. TDW θ frequency is also not constant even within a genotype but tightly modulated across circadian and behavioral contexts (SI Text and Fig. S5). Such θ rhythm variations may affect process-S dynamics. Driving S by the number of θ oscillations per unit of time, rather than by the total time in TDW, may address these issues. Second, TDW’s γ oscillatory component may allow refining the TDW definition or process-S equations. Third, Hcrt^ko/ko mice, similar to narcolepsy patients, tend to transition into PS prematurely (Fig. S6) and express more PS than WT in dark phase (Fig. S2). High selective pressure for PS was shown to suppress SWS δ content (27), and a PS contribution to process-S dynamics was recently proposed (28). Therefore, the impact of PS, as well as quiet waking, in SWS δ modeling needs to be explored.

Fig. S5.

TDW θ peak frequency is sharply modulated across time of day and behavioral contexts. (A) Heatmap representing TDW EEG spectral profile dynamics over the 3-d recording in Hcrt^+/+ (WT; n = 7) and Hcrt^ko/ko (KO; n = 8) mice. Contour lines connect EEG frequency (ordinate) and time (abscissa) pairs of equal relative EEG power density at 0.15% increments. EEG spectra were normalized as in Fig. 2A. Warmer colors denote higher power density. Blue lines connect mean θ peak frequency in TDW calculated for each time interval. θ peak frequency was recorded for each 4-s TDW epoch within a given time interval and then averaged within each mouse before constructing genotype means. (B) EEG spectra of TDW (Left) and PS (Right) in Hcrt^+/+ and Hcrt^ko/ko mice for 12-h light (dark yellow) and dark (gray lines) phases of 2 baseline days. EEG spectra were normalized as in A and Fig. 2A. (C) Boxplot summary of θ peak frequency in TDW (Left) and PS (Right) for 12-h dark (gray) and light (dark yellow) phases in baseline. Note the different abscissae for TDW and PS. Bar-width span, 10th and 90th percentiles; error bars, 5th and 95th percentiles. Red and black vertical lines within bars indicate mean and median, respectively. TDW θ peak was highly significantly shifted to faster frequencies in dark phase compared with light phase, in both genotypes (dark asterisks). However, the shift was smaller in Hcrt^ko/ko mice (two-way rANOVA; factor genotype, P = 0.0089; LD, P < 0.0001; interaction, P = 0.027), and TDW θ in dark phase remained slower in Hcrt^ko/ko than in Hcrt^+/+ mice (orange asterisk). In PS, a light-to-dark right shift in θ frequency was observed only in Hcrt^ko/ko mice (two-way rANOVA; factor genotype, P = 0.95; LD, P = 0.0016; interaction, P = 0.0059). PS θ peak frequency did not differ between genotypes. Orange asterisk indicates genotype differences (t test, P < 0.05); black asterisks indicate dark-to-light differences (paired t test, P < 0.05). Note that θ peak frequencies in C derive from assessing individual 4-s epochs scored as TDW (or PS) for their θ peak, followed by averaging. Values can slightly differ from θ peak frequencies determined by assessing average EEG spectra (as shown in B). For the same reason, blue curves in A (derived from assessing individual time intervals in TDW for θ peak, and then averaging) do not exactly follow the “crest” of average EEG spectra.

Fig. S6.

Latency to SWS and to PS in Hcrt^ko/ko and Hcrt^+/+ mice following a 6-h SD. (A) Time from the end of SD to the first PS episode (PS onset; Left) and to the first consolidated SWS bout (SWS onset; Middle), and PS onset relative to SWS onset (Right). See Methods for SWS onset definition. Vertical bars (KO, black; WT, gray) denote mean onset values (±1 SEM). SWS onset latency does not differ between genotypes, as was the case for EEG δ power during recovery SWS (Fig. 1A), both suggesting that homeostatic SWS pressure following SD is not affected by the mutation. In contrast, PS-onset latency is greatly reduced in Hcrt^ko/ko mice, to such an extent that PS occurs before the first consolidated SWS bout. Orange asterisks indicate significant genotype differences (t test, P < 0.05); black asterisks indicate differences from SWS onset (paired t test, P < 0.05). (B) The duration of this first PS episode was twice as long in Hcrt^ko/ko than in Hcrt^+/+ mice, albeit not significantly so (t test, P = 0.07; Left), and time spent in SWS between SD end and PS onset was greatly reduced (Right).

Altogether, our model proposes that sleep intensity, as reflected in δ oscillatory activity, which remains the best measure we have of the physiological need for sleep, depends on time spent in TDW, a state associated with the highest motivational drive (29) and neuronal activity-dependent plasticity (4). Although a waking θ/SWS δ correlation was reported by us and others (4–6), we here show SWS δ to be primarily determined by a discrete, fairly homogeneous, behavioral state. These insights may be relevant to explaining differences among various SD methods in eliciting EEG δ power rebound in recovery sleep. We speculate that failure to induce a pronounced δ rebound in some cases (30) may be due to the fact that, although these protocols successfully maintain animals awake, they do not induce robust θ activity and TDW state, as we show our gentle handling protocol to do.

Behavioral, Cellular, and Molecular Correlates of the TDW State.

What underlies the relation between the TDW mode of waking and SWS δ activity? In rodents, θ activity generally refers to 5- to 10-Hz oscillations of uncertain functional unity. Dynamics of θ in waking are complex and influenced by the animal’s behavior. In quiet wake and automatic behaviors, low-frequency (∼5.5-Hz), broad-bandwidth θ reflects sleep propensity in rodents (31, 32) and humans (33). In voluntary behaviors, such as explorative locomotion, foraging, and alert immobility, a higher-frequency (7- to 10-Hz), narrow-bandwidth θ rhythm of hippocampal origin appears (29, 34). Thus, θ can both reflect homeostatic sleep drive (in quiet wake) and contribute to it (in active wake). Whether these activities are related is unknown, but it is the latter we define as a discrete behavioral state (TDW) predicting sleep intensity. Interestingly, the behaviors typically associated with TDW, such as spontaneous exploration (35) or active waking in head-restrained rodents (36, 37), witness the highest Hcrt cell-firing rate. Lack of Hcrt signaling in brain circuits of actively exploring KO mice likely contributes to TDW episodes’ premature ending. KO mouse cataplexies likewise are triggered in motivated activities associated with high θ/γ EEG power (21), and their incidence parallels TDW state prevalence in early dark phase. This suggests that Hcrt release is needed as mice enter TDW, not only to sustain θ/fast-γ network activity but also to preserve the state’s functional integrity, including robust coupling to muscle tone.

We found that, although movement counts and TDW prevalence are globally highly correlated in both genotypes, the two variables uncouple during SD, suggesting TDW can be expressed in the absence of overt locomotion. TDW/LMA uncoupling is consistent with rat studies showing that prominent hippocampal θ typically accompanies vigorous locomotion but can exist in the resting animal preceding or following a running bout (34, 38) or under high motivational drive (29). Thus, TDW can occur in the absence of LMA, and LMA in the absence of TDW also likely occurs, as automatic locomotor activity such as wheel running was found to be associated with reduced cortical firing rates compared with exploration (39). Therefore, our data argue that TDW is not a simple readout of motor activity but a more sensitive marker of motivated waking.

Whether and through which mechanisms TDW is causally linked to SWS δ power hinge on issues of network homeostasis across behavioral states. If sleep is a response to waking cellular events that “use dependently” cause sleep need (7, 26, 40) to serve synaptic homeostatic and/or mnemonic sleep functions (41–43), TDW is expected to capture the homeostatic determinants of the “cost of waking.” In support of this, hippocampal θ activity is associated with heightened synaptic plasticity and memory-related potentiation, and long-term potentiation (LTP) in the hippocampus is optimally induced by θ rhythms resembling those seen in exploring animals (44). TDW may thus lead to θ-dependent increases in connectivity, which once in SWS result in enhanced δ synchrony (45, 46). Upon prolonged TDW, however, potentiation processes are expected to saturate, just as process-S dynamics themselves saturate. SD indeed is found to impair LTP capacity (47, 48) that only sleep can restore. Our molecular data support a role of TDW in synaptic plasticity, as Hcrt^ko/ko mouse lower TDW expression correlates with lower cortical expression of the plasticity-related transcripts Egr1/Zif268, Per2, and Bdnf. Consistent with prior studies (4, 49), Bdnf mRNA displays the highest correlation with EEG δ power across light-to-dark transitions and the two genotypes. Interestingly, KO mouse altered early dark-phase wakefulness also correlates with markedly up-regulated plasma corticosterone (Fig. S7).

Fig. S7.

Increased plasma corticosterone in Hcrt^ko/ko mice in the period surrounding dark onset and rise in activity. To further probe Hcrt^ko/ko mouse early dark phase for physiological correlates of their reduced TDW expression, blood plasma was collected and assayed for corticosterone, a hormone increasing with stress, and showing a sharp circadian regulation across that time of day. In WT C57BL/6J mice, plasma corticosterone peaks 1 h before dark onset (ZT11) and quickly wanes thereafter to reach a minimum 3 h after dark onset (ZT15) (77). Surprisingly, levels were markedly elevated in Hcrt^ko/ko mice, whereas time-of-day regulation appeared preserved. Hcrt^ko/ko mouse levels were 1.9-fold higher at ZT11 relative to Hcrt^+/+ littermates (KO 110.8 ± 11.7 ng/mL; WT 58.6 ± 7.1 ng/mL), and 2.4-fold higher at ZT15 (KO 30.7 ± 6.2 ng/mL; WT 12.7 ± 4.4 ng/mL; P < 0.0001 for overall genotype effect, two-way ANOVA). A 3-h SD intervention across ZT12 to 15 induced corticosterone to rise in both genotypes, with a 10-fold increase in Hcrt^+/+ mice (ZT15+SD relative to ZT15−SD) but only a 3-fold increase in Hcrt^ko/ko mice. This lower corticosterone response to SD may reflect reduced attention and behavioral adaptiveness, consistent with the diminished autonomic defensive response of KO mice when exposed to an intruder (24). It is noteworthy that, as for other variables, SD is seen to normalize Hcrt^ko/ko mouse phenotype also for plasma corticosterone, as levels no longer significantly differed in the two genotypes after SD. Vertical bars (KO, black; WT, gray) denote mean values (±1 SEM). Orange asterisks indicate significant genotype differences (t test, P < 0.05); black asterisks indicate differences among the three conditions within genotypes (t test, P < 0.05) [number of mice per condition (KO/WT); ZT11, n = 15/13; ZT15−SD, n = 12/14; ZT15+SD, n = 8/7].

TDW State Heterogeneity: Hcrt Loss Finely Alters the TDW Oscillatory Spectrum.

We found Hcrt^ko/ko mice to express a TDW state with normal θ power but slower frequency than WT mice. Although TDW has by definition an almost monochromatic (∼8-Hz) θ power spectrum, its in vivo expression is associated with other oscillatory activities, notably in the high-frequency range, that also get modified by Hcrt loss. In baseline dark TDW and the minute preceding cataplexy, waking EEG spectra are very similar, including in featuring a distinct slow-γ (40-Hz) peak that is more powerful in KO than WT mice. SD, which is highly enriched in TDW in both genotypes, however, lacks this slow-γ peak. This suggests that Hcrt-dependent, goal-driven behaviors, but not Hcrt-independent, SD-enforced ones, solicit slow-γ–associated events, which are exaggerated in mice lacking Hcrt, and perhaps in narcolepsy. Abnormally high β and slow-γ synchrony is associated with motor disorders, including Parkinson’s disease (PD). Depletion of striatal dopamine also causes amplified β (11- to 15-Hz) and slow-γ (40- to 53-Hz) activity in mice (22). The “antimovement” β band of PD patients correlates with impaired movement, and decreases with therapy and functional improvement. The enhanced slow-γ power of KO mouse baseline TDW may likewise be linked to motor deficits such as partial cataplexy. Hcrt signaling appears thus critical to limiting slow-γ activity in baseline dark-phase TDW, whereas under SD other arousal pathways sustain a TDW state largely lacking this activity. Supporting this, locus coeruleus norepinephrine (NE) cell stimulation, as likely occurs in SD, suppresses β (12- to 20-Hz) and slow-γ (20- to 40-Hz) activity in rats (50).

TDW stability thus seems to be mediated by context-dependent arousal pathways. Self-motivated TDW critically relies on Hcrt, whereas enforced TDW is largely Hcrt-independent and may partly rely on NE and threat-avoidance pathways. That TDW during SD is Hcrt-independent is consistent with studies showing that after 2-h SD, Hcrt cells become inhibited by NE (51). As sleep pressure builds up, Hcrt cell activity may thus fade and arousal increasingly rely on other neuromodulators and circuits, such as NE pathways. Hcrt-dependent behaviors are also thought to share an outreaching, positive emotional component (12, 52), as seen in cataplexy’s failed coupling of muscle tone to emotional arousal (53), whereas pain-avoidant behaviors appear less affected in Hcrt absence (52). Gentle handling during SD may also induce Hcrt-independent threat-avoiding pathways that preserve TDW stability, explaining TDW contextual dichotomy.

Hcrt Deficiency Affects the Spectral Quality of SWS.

We show that spontaneous waking is followed by SWS with a specific deficit in slow-δ in Hcrt^ko/ko mice. This is reminiscent of findings in rats depleted of cortical NE that likewise manifest a slow-δ deficit in SWS following 6-h SD (23). These rats, moreover, display reduced cortical expression of synaptic activity-associated genes in preceding wake (25) like our KO mice in baseline dark phase. What mechanism may underlie this selective slow-δ loss? Low NE or Hcrt neuromodulatory levels cause thalamocortical neurons to hyperpolarize, which Steriade and coworkers have shown to lead to faster δ oscillations (54), and hence a spectral right shift as seen in KO mouse SWS. Although these effects are robust, their impact on homeostatic recovery remains unclear.

Hcrt and θ/γ Network Activity.

What are the Hcrt-dependent processes whose lack in Hcrt^ko/ko mice leads to alter so profoundly the stability of the circuits generating θ and γ oscillations? Both anatomical and electrophysiological data support a link between Hcrt signaling and θ oscillations. Hcrt cells profusely innervate the medial septum and diagonal band of Broca in basal forebrain (BF), which are critical in pacing hippocampal θ, and also project to posterior cingulate and prefrontal cortices, which house cortical θ pacemakers (55). In BF, Hcrt excites both cholinergic and GABA septohippocampal neurons (56–59), two cell types important in θ regulation. The hippocampal θ rhythm during movement was recently shown to be orchestrated by networks of reciprocal connections between septal and hippocampal parvalbumin-positive (PV) GABA cells (60, 61). Hcrt is known to excite these septohippocampal PV cells and contribute to hippocampal θ enhancement and arousal. Investigating the role of Hcrt signaling in the regulation of these septohippocampal networks is warranted.

A direct role of Hcrt in cortical γ oscillations is supported by recent studies showing that Hcrt enhances glutamatergic input on, and firing rate of, mouse prefrontal cortex fast-spiking PV cells (62), a cell type whose firing is both necessary and sufficient for γ activity (63, 64). Hcrt^ko/ko mouse poor ability to sustain TDW may thus partly reflect the impaired activity of these prefrontal cells during motivated behaviors, which in WT mice correlate with Hcrt release. Supporting this hypothesis, local infusion of an HCRTR1 antagonist selectively reduces γ power (65), whereas infusion of Hcrt leads to improved accuracy in high-attention demanding tasks (66), known to correlate with fast-γ power (67). Moreover, cortical fast-spiking PV cells are critically involved in network stability and prevention of seizure propagation (64, 68). Cortical networks may thus be more vulnerable to uncontrolled hypersynchronies in Hcrt absence, which may explain the frontal high-amplitude paroxysmal θ bursts we and others reported in Hcrt-deficient mice during cataplexy and PS (21, 69) and in narcoleptic children during cataplexy (21). Failure of Hcrt neurotransmission in cortical fast-spiking PV cells may contribute to destabilizing θ/fast-γ network activities that underlie the TDW state. Thus, an Hcrt-responding cell type well-positioned to stabilize the TDW state in both BF and cortex is the PV GABAergic neuron.

What Is Sleepiness?

Heightened θ and fast-γ oscillations correlate with intensive exploration in rodents and mental concentration in humans (67), and are thought to provide temporal frames for neuronal excitability and spiking fluctuations, thus facilitating information transfer. TDW instability in Hcrt^ko/ko mice may thus weaken sensorimotor coding. Cognitive performance has been difficult to assay in narcolepsy due to the confounding effect of sleepiness. Sleep, however, remains restorative and phasic arousal is preserved in patients (70), and sleepiness-independent deficits were reported, including impaired perceptual encoding of stimuli, attention deficits in executive tasks, and impaired decision making (70–72). Neural substrates underlying sleepiness and subjective perception of sleep propensity while awake are largely unknown. Could TDW instability model aspects of sleepiness? Increased wake-to-SWS and SWS-to-wake transitions (15), and TDW fragmentation, may induce a feeling of sleepiness. It is worth noting in this context that the extent of the debilitating sleep attacks experienced by narcoleptic patients in a given day do not correlate with the preceding night sleep quality (73), emphasizing that, although commonly referred to as a sleep disorder, narcolepsy is primarily caused by weakened networks that sustain active waking, independent of sleep homeostatic processes and sleep amount.

Perspective.

The Hcrt^ko/ko mouse models two segregated types of waking, spontaneously motivated waking in early nocturnal phase and enforced waking, associated with diverging oscillatory activity profiles and sleep homeostatic weights. This stems from the two waking modes’ differential Hcrt dependence, where only the former depends on Hcrt to sustain the θ/fast-γ network activity that defines active waking, or TDW. The TDW variable allows quantifying active wake in distinct behavioral contexts, and successfully predicts δ power dynamics in Hcrt^+/+ and Hcrt^ko/ko mice, baseline and recovery SWS, by assuming that TDW drives the sleep homeostat. Sleep homeostatic drive is thus primarily caused by events representing only a third of waking time, and TDW bouts presumably frame in time the activity of the causal substrates for the elusive physiological need for sleep.

A recent study provided evidence supporting that synaptic homeostasis, specifically the firing rate rebound occurring in visual cortical neurons following monocular deprivation, occurs exclusively in active waking (19), in striking contrast to the “SHY” hypothesis, which holds synaptic homeostasis as sleep’s major function (41). A test of this contention may be to expose Hcrt^ko/ko mice to monocular deprivation and determine whether their reduced TDW expression correlates with a slower efficiency in restoring firing rates, relative to a WT control group. Moreover, TDW is distinguished from other states by its neuromodulatory signature, of which elevated Hcrt is a component, as Hcrt is maximally released in active waking (36) and by the θ/fast-γ oscillatory network activity we use to define it, and which in baseline nonredundantly depends on Hcrt. Hcrt signaling was shown to mediate synaptic plasticity in some contexts (74) and may also play a role in gating synaptic homeostatic events. Hence, testing Hcrt’s potential role in synaptic homeostatic processes is a prerequisite for interpreting the above experiment.

Because TDW is also likely associated with learning and related potentiation events (46), the TDW state would confine in close temporal association Hebbian potentiation events, synaptic homeostatic events that the former may contribute to causing to ensure network sustainability, and events determining sleep need, and thus potentially causally linked to SWS recovery functions. Elucidations of the mechanisms through which Hcrt signaling, in concert with other neuromodulators, affects the function of wake-active cortical networks in the healthy brain and neurological disorders are major goals of future research.

Methods

Mice.

All analyses are based on comparing littermate Hcrt^ko/ko (KO) and Hcrt^+/+ (WT) 3-mo-old males that were offspring of heterozygous Hcrt^tm1Ywa (Hcrt^+/ko) parents (75). Mice were bred at the 10th C57BL/6J-backcross generation. Recordings described here were analyzed for cataplexy in a separate study (21). Differential frontoparietal EEG/EMG (electromyogram) data were acquired and analyzed as described (76) (SI Methods for details).

All animal procedures followed Swiss federal laws and were preapproved by the State of Vaud Veterinary Office. At all times care was taken to minimize animal discomfort and avoid pain.

Theta-Dominated Wakefulness.

An algorithm was created that analyzes the EEG of each 4-s epoch scored as waking, and identifies those meeting TDW-defining criteria. If the frequency bin with the highest power density in the 3.5- to 15-Hz range is within the 6.5- to 12.0-Hz range, and the ratio of θ power in the TDW θ peak frequency ± 1-Hz range over total power across 3.5 to 45 Hz is above 0.228, then the epoch scores as TDW, if the three following criteria are met: (i) the epoch immediately preceding was scored as “W” or “TDW”; (ii) the epoch immediately following was not scored as SWS; and (iii) single-TDW epochs preceded and followed by three W epochs are excluded. The 0.228 θ-to-total power ratio was chosen because algorithmically determined TDW epochs using this threshold yielded the best match with epochs visually identified as active waking based on their stereotypical, regular EEG θ rhythmic signal, in two WT and one KO mice. Moreover, video-assessed intense behaviors that typically precede cataplexy (21), and exploration following cage change, correlated with TDW scoring. Effect of the choice of the 0.228 threshold on genotype differences in TDW expression is assessed in Fig. S3.

Process-S Simulation.

S is iteratively calculated in each mouse based on its 4-s-by-4-s behavioral state scoring sequence, assuming that it increases as a saturating exponential in wake and PS epochs according to S[t + 1] = UA − (UA − S[t]) × e^−dt/τi and decreases exponentially in SWS epochs according to S[t + 1] = LA + (S[t] − LA) × e^−dt/τd, where S[t] and S[t + 1] are consecutive S values, which vary between upper (UA) and lower (LA) asymptotes, with rate constants τi and τd, for S increase and decrease, respectively (5). The S simulation shown in Fig. 1A (Top) was performed using published τi, τd, UA, and LA parameter values (17). Subsequent simulations optimized the parameters in each mouse by searching for values yielding the best fit with measured δ power values. Fit was quantified as the least of the mean square of the differences upon running the simulation ∼200,000 times per mouse across the 3 d with a wide range of values for the four free parameters (Fig. 4B, Top). S value at t = 0 (S[0]) was estimated assuming that the 2 baseline days represent the steady state. For each four-parameter set, the simulation was run first for the two baselines with S[0] set at 148% (17). S average at the end of each of the two baselines was used as S[0] to initiate the final 3-d simulation. Optimization was based on 2 × 18 time intervals in baseline and 14 time intervals in recovery as the SWS δ power time course. This was repeated assuming S increases only in TDW and decreases only in SWS (Fig. 4B, Bottom). To draw the resulting time courses, process-S values were averaged at 15-min intervals.

Quantitative RT-PCR.

PCRs were performed using both custom-designed and commercial TaqMan primers on an ABI PRISM 7900 HT real-time thermocycler (SI Methods for details).

Statistics.

TMT Pascal Multi-Target5 software (Framework Computers Inc.) was used to process data, SigmaPlot version 10.0 (Systat Software Inc.) was used for graphics, and SAS version 9.2 (SAS Institute Software Inc.) was used for statistics. Genotype effect on sleep/wake distribution, EEG power, and time course was assessed using two- or three-way repeated-measures analysis of variance (rANOVA). Genotype and time-of-day effects on mRNA level and plasma corticosterone were evaluated by ANOVA. Significant effects and interactions were decomposed using post hoc Tukey’s honest significance and t tests. Statistical significance was set at P = 0.05, and results are reported as mean ± SEM.

SI Text

TDW θ Peak Frequency Is Sharply Modulated Across Time of Day and Behavioral Contexts.

To better understand the dynamic range of TDW’s EEG, we examined how its spectral content is modulated across time of day and behavioral contexts. Sharp modulations in TDW’s θ peak frequency, with an abrupt frequency increase after dark onset, are seen in both genotypes, and a more modest increase in early SD (Fig. S5A). These θ frequency dynamics are partly distinct from those in θ power, as θ power peaks during SD, and also partly distinct from LMA variations (compare Fig. S5A with Fig. S1B, Bottom). Plotting TDW spectra separately in light and dark phase (Fig. S5B, Left) further illustrates the dark/light (D/L) dichotomy, with a dark-phase θ frequency right shift present in both genotypes but less pronounced in KO mice (D/L difference; WT +0.57 ± 0.04 Hz; KO +0.43 ± 0.04 Hz; P = 0.027, t test; Fig. S5C, Left). Moreover, the genotype difference in TDW’s θ peak frequency, by which KO mouse TDW θ rhythm is slower (Fig. 3 C and D), is more pronounced in the dark. What causes these sharp TDW θ frequency variations is unclear. Rat studies showed that hippocampal θ frequency increases with locomotor behavior incidence and speed (29, 34). Circadian and sleep/wake–driven differences in locomotor activity, neuromodulatory tones, temperature, and other state-related variables with slower kinetics than those of behavioral state transitions may contribute to variations in a state’s EEG variables.

PS θ Peak Frequency Is Sharply Modulated Across Time of Day in KO Mice.

Surprisingly, a similarly pronounced right shift in θ peak from day to night is seen in PS, but only in KO mice (in dark 7.78 ± 0.06 Hz vs. in light 7.06 ± 0.04 Hz; Fig. S5B, Bottom Right and Fig. S5C, Right), confirming that this θ frequency increase is not necessarily linked to locomotion. This phenotype may be relevant to narcolepsy, as dark-phase PS is primarily a symptomatic trait in KO mice. Like narcoleptic patients, Hcrt^ko/ko mice display excess PS in active phase, reduced PS latency, and direct wake-to-PS transitions (21, 75) (Figs. S5 and S6). Light-phase PS, in contrast, is more akin to normal, SWS-ensuing, PS. Here we show that dark-phase PS is spectrally distinct from light-phase PS in narcoleptic mice. Furthermore, because θ oscillations in phasic PS are known to display higher amplitude and frequencies than θ oscillations in tonic PS (78), this observation, coupled with the increased total time spent in PS (Fig. S2), might indicate a selective phasic PS up-regulation in Hcrt^ko/ko mice in the dark phase.

SI Methods

Mice.

All animal procedures followed Swiss federal laws and were preapproved by the State of Vaud Veterinary Office. At all times care was taken to minimize animal discomfort and avoid pain. Mice were under 12-h light/12-h dark, with light onset at 0800 hours (Zeitgeber 0; ZT0) and water and food ad libitum.

EEG/EMG, Video, and Locomotor Activity Recording.

Electrodes for differential frontoparietal EEG and EMG recording were implanted as described (76). The parietal electrode tip effectively contacts the hippocampus, as cortical thickness is ca. 1 mm at that site. Recordings consisted of 2 “baseline” days, followed by a 6-h sleep deprivation (SD) initiated at light onset (ZT0) and performed by “gentle handling” (76), followed by an 18-h recovery period. Behavioral monitoring was performed using infrared video cameras allowing close frontal viewing. Locomotor activity was assessed at 1-min resolution using top-mounted passive infrared motion detectors (Visonic) and analyzed with ClockLab software (Actimetrics). EEG and EMG signals were amplified, filtered, analog-to-digital–converted, and stored at 200 Hz (EMBLA A10 and Somnologica-3; Medcare Flaga; Thornton). Each 4-s epoch was assigned a behavioral state score using published criteria (21). Analyses of state-specific EEG spectral profiles, the dynamics of SWS EEG δ power across the 3 d and at transitions to SWS, and waking EEG θ and γ power dynamics were performed as previously published (18, 21, 76, 79). Sleep-onset latency following SD was calculated as the time elapsed between SD end and the first SWS episode lasting ≥1 min and not interrupted by more than two 4-s epochs scored as wakefulness (18). PS-onset latency was defined as the time elapsed between SD end and the first 4-s epoch scored as PS.

EEG Power Spectral Density Analysis.

EEG signals were subjected to discrete Fourier transform to determine EEG power density spectra (0 to 90 Hz) for 4-s windows (Hamming function) with 0.25-Hz frequency resolution. Hardware (EMBLA A10) and software (Somnologica-3) were purchased from Medcare Flaga (EMBLA; Thornton). Mean EEG spectral profiles for each behavioral state and time interval were calculated using artifact-free, same-state–flanked 4-s epochs. To account for interindividual differences in absolute EEG power, power density in each frequency bin and for each state was expressed as a percentage of a reference value, calculated across the 2 baseline days for each individual mouse as the mean total EEG power across all frequency bins (0.75 to 47.50 Hz) and all behavioral states. This reference value was weighted so that for each animal each state contributed equally to the total EEG power (79). The total baseline EEG power values used as reference (100%) did not differ between genotypes (Hcrt^ko/ko 756 ± 81 vs. WT 744 ± 98 μV²; P = 0.92, t test).

Time-Course Analysis of δ Power in SWS.

To assess the dynamic changes in EEG power in the δ frequency range (i.e., EEG δ power, 1 to 4 Hz) for 4-s epochs scored as SWS, the time course of δ activity was performed as described (80). EEG δ power density was calculated by averaging power density in the frequency bins from 1 to 4 Hz. To account for individual differences in absolute EEG, δ power values were expressed as a percentage of the mean value reached across the last 4 h of the light (rest) periods (ZT8 to 12) (100%) when δ power is minimal in baseline, consistent with the notion that δ power reflects homeostatic sleep pressure, which is lowest at the end of the main sleep period. These reference δ power values did not differ between genotypes [KO 31.1 ± 3.9 vs. WT 30.5 ± 5.4 (μV²/0.25 Hz); P = 0.94, t test]. Only δ power values of epochs that themselves, as well as the two adjacent ones, were scored as artifact-free SWS were included in the analysis (representing 78% of all SWS epochs). To establish the time course, SWS δ power was averaged for each of 12 successive time intervals to which an equal number of SWS epochs contributed during the 12-h light periods, for 6 intervals during the 12-h dark periods, and for 8 intervals during the 6 h following SD.

Time Course of δ Dynamics at Transition to SWS.

Analysis of δ dynamics at transition to SWS was performed as described (79). Transitions into SWS were defined as four or more consecutive 4-s epochs scored as SWS preceded by eight or more 4-s epochs not scored as SWS (i.e., wakefulness or PS). δ power was then analyzed in epochs not scored as SWS in the minute preceding the transition and scored as SWS in the 2 min following the transition. Average transition curves were then constructed by aligning and averaging these 3-min windows first within animals and then across animals of each genotype. Values were expressed relative to the same individual δ power reference used for the 3-d time course of δ power.

Time-Course Analysis of θ and γ Power in Wakefulness.

Similar to the daily time course of δ power in SWS, the dynamics of changes in θ (6.0 to 9.5 Hz), low γ (γ-1; 32 to 45 Hz), and high γ (γ-2; 55 to 80 Hz) activity in waking EEG were calculated. The number of intervals per recording section was chosen according to the prevalence of wakefulness, namely 6 in the baseline light periods, 12 in the dark periods, 8 during the 6-h SD, and 4 during the recovery light period. EEG power density values in each frequency range were expressed as a percentage of the mean value reached at the end of the baseline light periods (ZT8 to 12). These reference values did not differ between genotypes [for θ: KO 6.02 ± 0.72 vs. WT 6.02 ± 0.67 (μV²/0.25 Hz), P = 1.0; for slow-γ: KO 0.46 ± 0.06 vs. WT 0.42 ± 0.07 (μV²/0.25 Hz), P = 0.67: for fast-γ: KO 0.15 ± 0.02 vs. WT 0.16 ± 0.03 (μV²/0.25 Hz), P = 0.75; t test].

To analyze the time dynamics of the entire EEG spectrum (0.75 to 90 Hz), power density in each of the individual frequency bins was analyzed in exactly the same fashion as described for the three frequency bands above, namely power density in each frequency bin was first expressed at its level reached during the last 4 h of the baseline light period before averaging over time intervals and individuals of each genotype. The results were visualized as a 3D “heatmap,” with larger changes in power density appearing as warmer colors.

Dynamics of TDW EEG Spectra Across the 3-D Recording.

Heatmaps were constructed to depict the TDW EEG spectra at different times across the 3 d with particular focus on the change in the θ frequency range. The spectra were expressed first as a percentage of total EEG power at baseline as explained above to reduce variance due to interindividual differences in signal power. In addition, the time course of TDW θ peak frequency (TPF) itself was calculated and overlaid on the heatmap. For this, TPF of each 4-s epoch scored as TDW was calculated for each 4-s epoch and averaged first over time intervals and then within and across animals of each genotype. Note that the frequency at which power density peaks in the average EEG spectrum does not necessarily correspond 1:1 with the average of TPF determined in individual 4-s epochs.

Quantitative RT-PCR.

Mice were killed by cervical dislocation within 30 min of the ZT time indicated (ZT11 or 15). Eleven- to 16-wk-old males were used (n = 7 to 9 per time point, condition, genotype; total n = 45 mice). Brains were removed rapidly, meninges were peeled off, olfactory bulbs and cerebellum were removed, and cerebral cortex was dissected, flash-frozen in liquid nitrogen, and stored at −80 °C. For the ZT15 time point, animals were killed and their brain was removed under red LED illumination (625 to 630 nm). Cortical RNA was isolated and purified using an RNeasy Lipid Tissue Mini Kit (Qiagen), followed by DNase treatment. RNA quantity was assessed using a NanoDrop ND-1000 spectrophotometer (Thermo Scientific) and quality-controlled using a Fragment Analyzer (Advanced Analytical Technologies). RNA was analyzed by quantitative RT-PCR (TaqMan) to assess steady-state mRNA levels of Arc, Bdnf, Egr1, Fos, Homer1a, Hspa5, and Per2, as well as three housekeeping genes, Eef1a1, Tbp, and Rps9. First, 700 ng of each total cortical RNA sample was reverse-transcribed using random hexamers and SuperScript II reverse transcriptase (Invitrogen). Next, the cDNA equivalent of 7 ng of total RNA was PCR-amplified in an ABI PRISM 7900 HT detection system (Applied Biosystems). To assay expression of Arc, Bdnf, Homer1a, and Per2, sequences of forward and reverse primers and probes were as described (77). For Egr1, Fos, and Hspa5, “best coverage” TaqMan gene assays (Life Technologies) were used. The effective amplification performance of each transcript assay was determined by PCR-amplifying a serial dilution of one of the WT mouse brain cDNA samples in parallel to the other samples. We performed three technical replicates of each PCR. Expression levels were calculated using the modified ddCt method from qBase (Biogazelle), corrected using the effective amplification per PCR cycle observed for each transcript assay, and normalized relative to the three housekeeping genes.

Plasma Corticosterone.

Mice were singly housed and habituated to manual handling for 7 consecutive days before the first blood collection. Each mouse was bled twice at a ≥20-h interval. Mice were briefly anesthetized using 4% isoflurane for 30 to 45 s, and a small tail incision was made. Blood was collected in a heparin-coated capillary tube and centrifuged at 1,500 × g, and plasma samples were stored at −80 °C. Corticosterone was quantified by an enzyme immunoassay (ELISA Kit; Enzo Life Sciences) according to the manufacturer’s instructions. For a second set of mice, animals were killed by cervical dislocation, followed by decapitation, and trunk blood was collected in heparin-containing tubes.