Melanopsin as a Sleep Modulator: Circadian Gating of the Direct Effects of Light on Sleep and Altered Sleep Homeostasis in Opn4−/− Mice

In a first set of experiments, the acute effects of light and darkness on sleep and the ECoG were investigated by exposing mice of both genotypes to a single, 1-h pulse of light administered 3 h after the habitual LD transitions; i.e., Zeitgeber time (ZT)15–16 (with ZT0 referring to the time of light onset) and to a 1-h dark pulse given 3 h after light onset (i.e., ZT3–4). Mice were otherwise kept under LD 12∶12 conditions, and at least 14 d separated any two recording conditions.

In Opn4^+/+ mice, the light pulse readily increased the amounts of both rapid eye movement (REM) and non-REM (NREM) sleep at the cost of wakefulness, whereas the same light pulse in Opn4^−/− mice failed to affect time spent in either behavioral state for the duration of the light pulse (Figure 1A). This lack of a response demonstrates that, at least at this time of day (see LD 1∶1 results below), melanopsin contributes significantly to the acute effect of light on sleep.

Light-to-dark transitions induce waking and alertness in nocturnal rodents. Accordingly, in Opn4^+/+ mice, a 1-h dark pulse induced an immediate increase in time spent awake (Figure 1B), and the ECoG during wakefulness showed a rapid and prolonged induction in ECoG theta (7–10 Hz) and gamma (40–70 Hz) activity (Figure 1C and 1D), the ECoG correlates of exploratory behavior and alertness in rodents [25],[26]. The dark pulse also induced waking in Opn4^−/− mice. Although their immediate response was somewhat delayed (compare genotypes for the increase in waking in the 5 min before and after dark onset in Figure 1B), the hourly amount of waking (and sleep) did not differ between genotypes (unpublished data). The changes in ECoG theta and gamma activity that follow the transition into darkness seemed, however, to be modulated by melanopsin because they were delayed by ca. 25 min (Figure 1C and 1D). Upon restoring the normal light condition 1 h later, Opn4^−/− mice stayed awake longer, consistent with the light-pulse results (Figure 1B and 1A).

The single pulses of light and darkness probed their effects on sleep and waking at specific times of day only. To obtain a more complete account of the genotype differences of these interventions, we maintained mice under a short, 1-h LD cycle (LD 1∶1) for a 24-h period. This lighting schedule was used first in the rat to investigate the direct effects of light on sleep across the circadian cycle [27]. Also under these conditions, 1-h periods of light induced sleep, and 1-h periods of darkness favored wakefulness. Irrespective of the effect of this ultradian LD cycle on sleep and wakefulness, a circadian modulation of the average levels of sleep and waking reached within the subjective 12-h light period and subjective 12-h dark period is preserved in wild-type mice (Figure 2A; Table 1). The protocol has the potential drawbacks that light given at any one hour might influence its effects during subsequent hours and that animals could entrain their ultradian sleep–wake organization to this schedule. We did not see evidence of the former effect as the relative effect of the LD alteration on time spent awake did not vary with time of day in wild-type mice (Figure 2A–2C), but after the initial three cycles (i.e., after ZT6), wild-type mice seemed capable of anticipating dark onset, because increases in wakefulness started preceding it (Figures 2B and S2).

Visual inspection of the hourly values of wakefulness reached under the LD 1∶1 protocol suggests that Opn4^−/− mice have a reduced response to light in the (subjective) dark period (Figure 2A); i.e., the time at which we and others [20],[21] administered the single 1-h light pulse (see above). This reduction was, however, not observed at other times of day, and the LD-induced changes in time spent awake did not differ from wild-type mice during the subjective light period (Figure 2A–2C). The capacity of the 1-h LD cycles to alter the sleep–wake distribution was further analyzed by aligning the subsequent 1∶1-h LD cycles (Figures 2B and S2). This time course analysis revealed that, as was observed after the single dark-pulse experiment (Figure 1B), Opn4^−/− mice were again slower in initiating sleep after light onset (Figure 2B). Analyses of variance confirmed that the time course of the LD-induced changes in wakefulness and NREM sleep varied significantly according to genotype and time of day (i.e., 12-h subjective light period versus 12-h subjective dark period; Figure 2B). Changes in the time course observed in Figure 2B were summarized by estimating the amplitude of sine waves of best fit for individual mice (Figure 2C; see Text S1 for details). This analysis also confirmed that genotype affected the light-induced changes in wakefulness only during the subjective dark period (Figure 2C). Especially between ZT15–21, the increase of wakefulness during the 1-h dark intervals compared to the 1-h mean value in the preceding light intervals was no longer significant in Opn4^−/− mice (+4.2±4.1 min; p = 0.38 and +26.2±5.3 min; p = 0.038; for Opn4^−/− and Opn4^+/+ mice, respectively; paired t-tests), and this increase differed between genotypes (one-way ANOVA; p = 0.020). It thus seems that only during this time melanopsin contributed significantly to the direct effects of light on wake and sleep duration.

As expected from our previous work and that of others [16],[17], a 1-h light pulse administered during the dark period induced c-Fos immunoreactivity in the SCN in both genotypes (two-way-ANOVA, light-pulse effect: p<0.001, followed by post hoc Fisher protected least significant difference [PLSD]: p<0,05); however, the induction was half that observed in Opn4^+/+ mice (two-way ANOVA, light×genotype interaction: p<0,001; light-pulse effect: Opn4^+/+ vs. Opn4^−/−, post hoc Fisher PLSD, p<0,05) (Figure 3). This reduction in c-Fos immunoreactivity has been functionally linked to the reduced ability to phase shift circadian rhythms in these mice. This does not, however, rule out the possibility that the reduced activation of SCN neurons could also contribute to the reduced direct effects of light on sleep and waking described here.

We further determined whether melanopsin conveys light information to the VLPO, in particular to the galaninergic sleep-active neurons, using double staining for c-FOS protein (immunohistochemistry) and for galanin mRNA (in situ hybridization). Whereas overall c-Fos immunoreactivity in the VLPO area was not significantly affected by the light pulse, in Opn4^+/+ mice, the percentage of galanin-containing (GAL) neurons coexpressing c-Fos did significantly increase compared to the control condition (i.e., darkness) (Figure 4). In contrast, in Opn4^−/− mice, GAL c-Fos-costained neurons were present at low levels in both lighting schedules (2%), and no induction of c-Fos in GAL neurons was observed, suggesting that at least at this time of day, melanopsin-containing RGCs contribute significantly to the effects of light on the activity of these sleep-active VLPO neurons (two-way ANOVA, light-pulse effect: p<0.05, light×genotype interaction: p<0,05; light-pulse effect: Opn4^+/+: p<0,05; Opn4^−/−: p = 0,23, post hoc Fisher PLSD).

The mouse is a nocturnal species that avoids light of higher intensity. Opn4^−/− mice are no exception and, similar to their littermate wild-type controls (Opn4^+/+), mostly sleep during the light period and are awake during the dark period of the LD 12∶12 cycle (Figure 5A; Table 1). Nevertheless, Opn4^−/− mice lose ca. 1 h of NREM sleep per day relative to Opn4^+/+ mice. This marked loss of sleep occurred during the 12-h light period exclusively (Figure 5B) and resulted in an attenuation of the diurnal distribution of sleep and waking (Table 1). This genotype difference could be due to a reduced capacity of light to induce sleep or to suppress wakefulness, an interpretation underscored by the results of the light- and dark-pulse experiments (Figure 1A and 1B) and the LD 1∶1 experiment (Figure 2A). Indeed, for both genotypes, the levels of sleep reached during the 1-h light intervals during the subjective light period of the LD 1∶1 schedule were similar to those reached during the light period of the LD 12∶12 schedule; i.e., Opn4^−/− mice slept less when light was present during the (subjective) light period of both the LD 1∶1 and LD 12∶12 (Table 1). Moreover, the sleep loss observed in Opn4^−/− mice was associated with an overall deficit in ECoG delta power during NREM sleep (see below).

The wakefulness present during the light period is organized in more or less regularly occurring bouts lasting on average about 20–25 min [28] (Figure S1; Table S1). To examine in more detail the mechanism underlying the differences in sleep time during the light period, we analyzed these spontaneous waking bouts. In Opn4^+/+ mice, after 20 min of sustained wakefulness, waking values reverted quickly to the low values characteristic of the light period (Figure 5C). In contrast, in Opn4^−/− mice, these waking bouts lasted on average 11 min longer (Table S1), and values remained above wild-type levels for 30 min (Figure 5C). The similarity between these results and the results observed after the single dark pulse (Figure 1B) suggest that the reduced capacity of light to suppress wakefulness and to induce sleep in Opn4^−/− mice also contributed to an overall reduction in sleep time during the light period.

Quantitative analysis of the ECoG during the three behavioral states revealed an increase in ECoG power density in the theta and gamma frequency bands in Opn4^−/− mice compared to in Opn4^+/+ mice (Figure 6). These differences were present both during the light and dark periods of LD 12∶12 (unpublished data). Such ECoG changes during NREM sleep are typically associated with reduced sleep quality [29], whereas during wakefulness, ECoG activity in these bands is associated with exploratory behavior, alertness, and cognition [25],[26],[30], and is thus markedly increased during waking compared to NREM sleep [31]. Genotype effects on theta activity could be analyzed in more detail in REM sleep because theta oscillations are especially prevalent and regular during this state [32]. Theta peak power during both REM sleep and wakefulness was nearly doubled in Opn4^−/− mice compared to controls (Figure 6; Table S2). The frequency of the theta oscillation during REM sleep was faster in Opn4^−/− mice compared to Opn4^+/+ mice during the light period, but not during the dark period. In mice, theta frequency during REM sleep is usually slower in the light compared to the dark period [32] (Table S2). In Opn4^−/− mice, no evidence for such dark–light slowing was, however, found (Table S2), suggesting that the melanopsin pathway can directly or indirectly modulate hippocampal activity.

Curtailing sleep time usually results in an increased need for sleep. It is therefore surprising that in Opn4^−/− mice, levels of ECoG delta power during NREM sleep, a reliable correlate of sleep need, were generally lower while less time was spent in NREM sleep compared to Opn4^+/+ mice. This reduction in delta power was especially pronounced in the dark periods of the LD 12∶12 schedule (Figure 7A). The same difference was observed during the subjective dark period under the LD 1∶1 schedule (Figure 7B) and thus did not depend on the presence or absence of light. Moreover, this reduction was not associated with an increased fragmentation of sleep (Figure S3) that could have interfered with the expression of delta oscillations.

The unexpected reduction of delta oscillations observed in melanopsin-deficient mice might result from alterations in the properties of the sleep homeostat. We tested the possibility that the build-up of a pressure for sleep when animals are awake occurs at a slower rate in a sleep deprivation experiment. Mice of both genotypes were kept awake by gentle handling for 6 h starting at light onset (ZT0–6). The level of delta power reached during NREM sleep immediately following the sleep deprivation was significantly lower in Opn4^−/− mice as compared to wild-type animals (Figure 7C), suggesting that in the absence of melanopsin, the dynamics of the sleep homeostat are altered.

A main finding of our study is the result of the LD 1∶1 experiment that revealed that time of day modulated the acute effects of light and dark only in Opn4^−/− mice. During a 7-h period of the subjective dark period, Opn4^−/− mice did not respond to the ongoing LD alterations with respect to sleep–wake induction. During this time, we and others [20],[21] applied the single 1-h light pulse that has led to the conclusion that melanopsin mediates the direct effects of light. Our data show that at other times of day, melanopsin is not necessary for mediating these acute light effects on sleep duration.

Because light perception in Opn4^−/− mice depends solely on rod–cone photoreception, it can be argued that the inability to respond to light at specific times of day results from circadian changes in rod–cone sensitivity. In intact mice, no time-of-day effect was observed in the response to the LD 1∶1 schedule, suggesting that melanopsin compensates for circadian changes in rod–cone sensitivity. Such interaction between the two photosensitive systems could take place in melanopsin-containing RCGs since these cells integrate rod–cone input important for mediating nonvisual light information to the brain [10]. Circadian modulation of retinal output has been described, and for example, variations in rod–cone electrical coupling were shown to lead to higher rod–cone photosensitivity during the dark period [36]. In contrast, another study showed higher cone sensitivity during the (subjective) light period, a circadian difference that required the presence of melanopsin [37]. It is unknown whether such effects exist also for the nonvisual, acute effects of light on sleep. Melanopsin protein levels are also known to vary across the day due to both circadian- and light-dependent influences reaching their highest levels during the light period and lowest during the dark period [38]. Assuming that differences in protein levels translate into differences in (blue) light sensitivity further complicates the issue and calls for further investigation.

We want to emphasize here that the above-described effects concern the acute effects of light observed at the transitions between lighting conditions. We presented evidence that melanopsin also affects several aspects of sleep during the sustained 12-h light period of the LD 12∶12 schedule, including the spectral composition of the ECoG, the frequency of theta oscillations during REM sleep, and waking bout duration. The genotype differences in the ECoG activation during the 1-h dark pulse was also observed during this period of the day.

Although circadian gating already occurs at the level of the retina [36],[39],[40], the 50% reduction in the activation of SCN neuronal activity after a light pulse in Opn4^−/− mice may also be relevant for the circadian variation of the direct effects of light in these mice. As SCN functionality is conserved in the absence of melanopsin [9], a potential role of the SCN in also mediating the acute, noncircadian effects of light could explain the time-of-day-dependent variation in light sensitivity. The SCN is well placed to relay such activation since it influences key areas for sleep–wake control, such as the VLPO.

Besides the SCN, other anatomical targets of melanopsin-containing RGCs have been identified [22],[24],[41]. Among the various projection areas, the VLPO is a good candidate for relaying light information to sleep–wake neuronal networks. The VLPO has been implicated in the induction of NREM sleep [42], and galanin- and GABA-releasing neurons within the VLPO have been functionally characterized as a cluster of sleep-active neurons. These sleep-active neurons are thought to actively promote sleep by inhibiting the ascending arousal systems [43],[44]. A regional analysis of c-Fos expression within the VLPO area was conducted by Lupi et al., 2008, by real time PCR from punched tissues [21]. This analysis revealed a 2-fold increase in c-Fos following a 1-h light pulse in wild-type mice that was absent in Opn4^−/− mice. Using a different method (i.e., counting the number of c-FOS immunoreactive cells in the VLPO), we found little evidence for a light-induced general activation of the VLPO area. Only the costaining of galanin and c-Fos used in our study allows for an evaluation of transcriptional changes specific to those VLPO neurons that could mediate the sleep-promoting effects of light. Here, we provide the first evidence that light indeed activates galanin-containing cells in the VLPO. This light-induced c-Fos immunoreactivity was absent in Opn4^−/− mice. This anatomical result is consistent with the complete absence of sleep–wake changes during the 1-h light pulse at this time of day, suggesting that the VLPO might play a role in mediating these effects of light on sleep. Our findings thus suggest that light directly impinges on VLPO sleep-active neurons thereby shifting the balance of the reciprocal inhibitory interaction towards sleep promotion and arousal inhibition.

The rapid induction of ECoG activity in both the theta and gamma frequency bands during the dark pulse in Opn^+/+ mice is consistent with the alerting effects of light-to-dark transitions in a nocturnal species. Theta oscillations accompany exploratory behavior [45] and are also involved in long-term potentiation (LTP) and learning [46]–[48]. Gamma activity accompanies a wide variety of cognitive processes, including perceptual processing, attention, arousal, object recognition, and in humans, language perception [49]. As in the present study, gamma activity is often associated with the presence of theta oscillation in rodents [50]. The downstream events initiated by melanopsin have been suggested to affect LTP as well as performance in learning and memory tasks [51]. In humans, nonvisual responses related to alertness and cognition are associated with changes in regional brain activity detected by positron emission tomography (PET) or functional magnetic resonance imaging (fMRI) [52],[53]. Application of blue light during a working memory task induces specific brain activity changes within a time frame of seconds [7].

ECoG delta power is widely used to reflect a sleep homeostatic process because its level monotonically increases with wake duration and decreases during NREM sleep [54]. The reduction of ECoG delta power in Opn4^−/− mice was pronounced and present during both the LD 12∶12 and LD 1∶1 schedules. This reduction was all the more surprising because NREM sleep time was reduced in these mice, and reduced sleep time is usually associated with increased delta power. The unexpected reduction of delta oscillations in melanopsin-deficient mice might result from alterations in the properties of the sleep homeostat. One possibility is that NREM sleep may have been more efficient in reducing sleep need in Opn4^−/− mice. However, by comparing the ECoG spectral profiles during NREMS, evidence to the contrary was found; the increase in theta and gamma activity combined with reduction of delta activity in Opn4^−/− mice indicate that NREM sleep seems less profound and, if anything, likely to be less efficient in reducing sleep need and ECoG delta power [55]. The other possibility is that the buildup of a pressure for sleep when animals are awake occurs at a slower rate. We confirmed this possibility in a sleep-deprivation experiment; the level of delta power reached during NREM sleep immediately following the sleep deprivation was significantly lower in Opn4^−/− mice as compared to wild-type animals. In fact, delta power levels reached after sleep deprivation in Opn4^−/− mice were as low as the lowest levels observed in a panel of six inbred strains of mice [56]. This reduced compensatory response to sleep loss suggests that indeed the need for sleep increases at a slower rate in Opn4^−/− mice. The experiment also demonstrates that the reduced delta power is not specific to lighting condition or time of day.

Although a role for melanopsin in sleep homeostasis is unexpected and not easily reconciled with current hypotheses on sleep homeostasis, the modulation of VLPO sleep-active neurons by melanopsin-containing RGCs could hint to a possible mechanism. Sleep-active neurons in the VLPO are thought to be a neuronal substrate of sleep homeostasis because more than 50% of galanin-containing neurons express c-Fos during recovery sleep after a sleep deprivation [57]. A final consideration that applies to all studies using noninducible loss-of-function mutants is that the altered relationship between the sleep–wake distribution and ECoG delta power reflects a developmental effect.

Our results provide evidence that melanopsin-containing RGCs contribute to the noncircadian, nonvisual direct effects of light on sleep and the ECoG correlates of alertness and cognition. Melanopsin's contribution to the acute affects of light on sleep duration was however limited to a ca. 7-h time window. Our findings suggest that the acute photic sleep promotion stems, at least partly, from a stimulation of VLPO sleep-active neurons, which in return, would lead to an inhibition of the arousal systems. Apart from these direct effects of light observed at the transitions between lighting conditions, the daily loss of NREM sleep over the 12-h light period in Opn4^−/− mice provides evidence that melanopsin also modulates the expression of sleep under sustained periods of light exposure. If confirmed in humans, our observations concerning the time-dependent effects of melanopsin's contribution to the acute effects of (blue) light as well as the effects of sustained light exposure will have applications for the clinical use of light therapy as well as for 24-h patterns in luminance [58]. Finally, the discovery that the homeostatic regulation of sleep need can be affected by a photopigment is intriguing and represents a novel concept in the field of sleep regulation.

All experiments were performed on adult male Opn4^−/− mice and wild-type littermates (as controls), and carried out in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals as well as local veterinary office and use committees at Stanford University (for details and genotyping, see Text S1 and Ruby et al. (2002) [17]).

The methods concerning the recording and analysis of the ECoG in mice are described in detail in Text S1 and elsewhere [32],[59]. Briefly, ECoG and electromyogram (EMG) recordings from mice implanted with a classical set of electrodes, were collected (using commercial hardware, EMBLA, and software, Somnologica-3) for the following conditions: (1) two continuous baseline days under LD 12∶12 cycle; (2) a 1-h light pulse (white fluorescent tubes, Philips F32T8/TL741 Hi-Vision 40 W, 4,100 K [Philips Lighting], broad spectrum, mainly 400–800 nm) administered 3 h after dark onset (ZT 15), (3) 1-h dark pulse administered 3 h after light onset (ZT 3), (4) a 1∶1 LD cycle for 24 h, and (5) 6-h sleep deprivation starting at light onset. The behavior in each 4-s epoch was classified as waking, REM sleep, or NREM sleep based on the ECoG and EMG signals according to standard criteria [32]. The ECoG signal (analog-to-digital converted) was subjected to discrete Fourier transform (DFT) yielding power spectra at 0.25-Hz resolution. For each state, an ECoG spectral profile (0 to 90 Hz) was constructed by averaging all 4-s epochs scored as that state.

Galanin ISH and c-Fos immunostaining on frozen sections with or without (control condition) a prior 1-h light exposure was carried out as described previously [60],[61] and are detailed in Text S1. Staining of GAL mRNA was chosen instead of staining the protein because it has been shown in mice that cell body staining of galanin-containing neurons by IHC is only reliable in colchicine-treated animals. In Opn4^+/+ and Opn4^−/− mice (n = 6 of each genotype), identification of c-Fos immunoreactive cells and sacrifice of animals occurred at the conclusion of a 1-h light pulse as well as without a light pulse as a negative control. Sleep was recorded in these animals, several weeks before, under a light pulse identical to the light pulse administered the day of the perfusion to confirm that their response to light was similar to those of the whole group. Before ISH series of 18-µm-thick sections were pretreated by an antigen retrieval procedure [62]. The mouse galanin probe (National Center for Biotechnology Information [NCBI] BC044055↗, covering a 716-base sequence of the GAL prepro-mRNA) was used in a dilution of 1∶500. The ISH protocol preceding the immunohistochemical protocol was identical to the procedure described previously [60],[61]. GAL mRNA was visualized using a horse radish peroxidase (POD)-labeled sheep-anti-digoxiginin antibody (Roche 1207733, diluted 1∶200), and Alexa-tyramide 488 (Molecular Probe; diluted 1∶100). Hereafter, sections were incubated with a rabbit anti-c-Fos antiserum (c-Fos antibody dilution: 1∶500; code no: 9412, [23] and visualized by Alexa568-conjugated goat anti-rabbit antibody (Molecular Probe; diluted 1∶1,000). Counting of GAL- and c-Fos-expressing neurons in the VLPO and SCN was conducted using a confocal microscope (Zeiss LSM 510; Brock and Michelsen) equipped with appropriate filter settings for detecting Alexa488 and Alexa568 was used. The quantification method is detailed in Text S1.

Differences in sleep amounts and quantitative EEG variables were determined by single- or multiple-way ANOVAs, followed by post hoc t-tests if 5% significance levels were reached. The differences in number of c-Fos, Gal; or c-Fos+Gal neurons were assessed by two-way (light and genotype conditions) ANOVA, followed by post hoc Fisher PLSD.