Dim light at night disturbs the daily sleep-wake cycle in the rat

Exposure to light at night (LAN) is associated with insomnia in humans. Light provides the main input to the master clock in the hypothalamic suprachiasmatic nucleus (SCN) that coordinates the sleep-wake cycle. We aimed to develop a rodent model for the effects of LAN on sleep. Therefore, we exposed male Wistar rats to either a 12 h light (150–200lux):12 h dark (LD) schedule or a 12 h light (150–200 lux):12 h dim white light (5 lux) (LDim) schedule. LDim acutely decreased the amplitude of daily rhythms of REM and NREM sleep, with a further decrease over the following days. LDim diminished the rhythms of 1) the circadian 16–19 Hz frequency domain within the NREM sleep EEG, and 2) SCN clock gene expression. LDim also induced internal desynchronization in locomotor activity by introducing a free running rhythm with a period of ~25 h next to the entrained 24 h rhythm. LDim did not affect body weight or glucose tolerance. In conclusion, we introduce the first rodent model for disturbed circadian control of sleep due to LAN. We show that internal desynchronization is possible in a 24 h L:D cycle which suggests that a similar desynchronization may explain the association between LAN and human insomnia.

targeted hypothalamic manipulations, and electrophysiological studies as compared to the mouse 5,26 . Therefore, in the present study we investigated the effects of 5 lux dim light at night on sleep-wake behavior and energy metabolism in Wistar rats.

Results
Vigilance states and sleep architecture. To investigate the effects of light at night on sleep-wake behavior we exposed rats to either a 12 h light (150-200 lux):12 h dark (LD) schedule or a 12 h light (150-200 lux):12 h dim white light (5 lux) (LDim) schedule and performed EEG/EMG recordings. LDim strongly diminished the amplitude of the daily rhythm in all three vigilance states. Introduction of LDim decreased the time spent awake in the first dim phase compared to the baseline dark phase, with a further decrease over the following days. LDim gradually increased the time spent awake in the light phase (Phase P < 0.001, Day P = 0.136, Interaction P < 0.001; Fig. 1). Specifically, LDim reduced the duration of waking episodes during the dark (dim) phase and increased the duration of waking episodes during the light phase (Supplementary Figure 1). LDim also increased the time spent in NREM sleep in the first dim phase compared to the baseline dark phase, with a further increase over the days. LDim gradually decreased the time spent in NREM sleep in the light phase (Phase P < 0.001, Day P < 0.001, Interaction P < 0.001; Fig. 1). Specifically, LDim increased NREM sleep episode duration in the dark (dim) phase and reduced NREM sleep episode duration in the light phase (Supplementary Figure 1). Finally, LDim directly increased the time spent in REM sleep during the first dim phase compared to the baseline dark phase, and gradually decreased the time spent in REM sleep during the light phase (Phase P = 0.002, Day P = 0.003, Interaction P = 0.002; Fig. 1). The increase in REM sleep during the dark (dim) phase was mainly due to increased REM sleep episode numbers (Supplementary Figure 1). In LD, the daily rhythm in the vigilance states did not significantly change over the 14 days period (Fig. 1). Total 24 h sleep time was not significantly affected by LDim (LD: 43.8 ± 1.7%, 14 days LDim: 47.2 ± 1.0%, P = 0.293).
In order to investigate homeostatic sleep drive, we analyzed the rhythm of slow wave activity (SWA, EEG power density between 0.75-4.0 Hz). SWA showed a clear day/night rhythm with a peak at the beginning and a trough at the end of the light phase. LDim gradually decreased the amplitude of the SWA rhythm (Day P = 0.001; Fig. 2a).  ) is repeatedly plotted to facilitate comparison with subsequent days (closed circles). Registrations on day 0 and day 1 started at ZT12, recordings on day 7 and day 14 started at ZT0, but all data are plotted from ZT0 for convenience. Data are represented as mean ± SEM. If a significant main effect was detected, post hoc tests for differences from baseline were performed with student's t-test and indicated by asterisks (P < 0.05) or double asterisks (P < 0.001).
When analyzing the changes in power density of all frequencies between 0.25 and 25.0 Hz within the NREM sleep EEG, we observed reduced power in the [16][17][18][19] Hz frequency domain in the course of 14 days LDim (P < 0.05, after significant Interaction P < 0.05). Therefore we analyzed the rhythm of this frequency domain in more detail. The [16][17][18][19] Hz frequency domain showed a pronounced daily rhythm at baseline with a peak at the beginning of the dark phase. LDim strongly reduced the peak and the amplitude of this 16-19 Hz rhythm (Day P < 0.001; Fig. 2b).

Circadian locomotor activity.
To further investigate the effect of LDim exposure on the circadian timing system, we exposed male Wistar rats to 10 days LD followed by 30 days LDim and 20 days constant darkness (DD). Subsequently, the animals were re-entrained during 30 days to LD and finally the animals were released into DD for 10 days (circadian experiment A). LDim reduced the strength of the 24 h rhythm, and induced a second, interacting, free running rhythm with a period of 25.1 ± 0.0 h (see arrow in Fig. 3b). This interacting free running rhythm was observed in all animals exposed to LDim. After transfer from LDim to DD, rhythmic strength remained low in the first ten days with a period similar to the second rhythmic component in LDim, followed by an increase in rhythmic strength in the following 10 days (Fig. 3 and Table 1). Three out of 8 animals were virtually arrhythmic in the first days after transfer to DD (in these animals no significant period could be detected in the first 7 days after release in DD) and showed an improvement of the rhythm in the course of DD. After re-entrainment to LD and transfer from LD to DD all animals showed a free running period of 24.2 ± 0.1 h in DD.
As a control, we verified whether the occurrence of two interacting free running rhythms was specific for LDim and not due to continuous exposure to minimally 5 lux, by exposing Male Wistar rats consecutively to at least 10 days LD, 16 days constant dim light (DimDim), 22 days constant light (LL), 32 days DimDim, 14 days LD, and 16 days DimDim (circadian experiment B). DimDim exposure induced period lengthening to 24.9 ± 0.3 hr, but in contrast to LDim exposure, no dual rhythms were observed (Table 1, Supplementary Figure 2). LL exposure induced arrhythmic behavior and subsequent DimDim exposure did not restore rhythmicity. After re-entrainment to LD and re-exposure to DimDim, again period lengthening was observed.
SCN clock gene expression. In order to investigate the representation within the SCN of the two rhythmic components observed in locomotor behavior, we analyzed SCN clock gene expression at the time when the two rhythmic components in behavior were aligned or misaligned. Per1 expression in the SCN showed a significant diurnal variation with higher expression at ZT6 compared to ZT18 (Time P = 0.004, Group P = 0.137, Interaction P = 0.495). Post hoc tests showed that the effect of Time was only significant in the LD group (P = 0.025), but not in the LDim-aligned (P = 0.272) and LDim-misaligned (P = 0.136) groups (Fig. 4). There was no difference in dorsal/ventral ratio between ZT6 and ZT18 in misaligned animals (P = 0.324).  [16][17][18][19] Hz activity compared to baseline. Data are expressed as percentage of the 24-h average at baseline. Registrations on day 0 and day 1 started at ZT12, but all data are plotted from ZT0 for convenience. Data are represented as mean ± SEM. If a significant main effect was detected, post hoc tests for differences from baseline were performed with student's t-test and indicated by asterisks (P < 0.05) or double asterisks (P < 0.001). Arntl expression in the SCN showed a significant diurnal variation in antiphase to Per1, with lower expression at ZT6 compared to ZT18 (Time P = 0.027, Group P = 0.574, Interaction P = 0.145). Post hoc testing showed that the diurnal variation nearly reached significance in the LD group (P = 0.058), but not in the LDim-aligned (P = 0.130) and LDim-misaligned (P = 0.899) groups (Fig. 4). There was no difference in dorsal/ventral ratio between ZT6 and ZT18 in misaligned animals (P = 0.176).

Energy metabolism and body weight.
Previous experiments in mice showed that dim light at night can influence energy metabolism 18 . Therefore, we investigated the effects of LDim compared to LD on food intake, energy expenditure, and body weight, both on a chow diet and on a high fat diet (HFD), with metabolic cages. On a chow diet, LDim induced a decrease of the amplitude of the daily rhythm of food intake from 84 ± 2% dark phase food intake at baseline to 59 ± 2% dark phase food intake in week 7. This decrease was much stronger compared to the amplitude decrease in LD from 83 ± 2% dark phase food intake at baseline to 77 ± 3% dark phase food intake in week 7 (Light schedule P = 0.426, Week P < 0.001, Interaction P = 0.026). LDim also induced a decrease in the daily rhythm of energy expenditure compared to the LD animals (Light schedule P = 0.212, Week P = 0.001, Interaction P = 0.003) (Supplementary Figure 3). Animals exposed to LD showed a small increase over time of total 24-h food intake compared animals in LDim (Light schedule P = 0.927, Week P = 0.002, Interaction P = 0.025), but total 24-h energy expenditure did not differ between LDim and LD (Light schedule P = 0.908, Week P < 0.001, Interaction P = 0.084) (Supplementary Table 1).
The increase in body weight did not differ between LDim and LD, neither on a chow diet nor on a HFD (Fig. 5 and Supplementary Table 2). White adipose tissue (WAT), adrenal, and thymus weights were not different between LDim and LD on either a chow diet or a HFD (Supplementary Table 2).
Glucose tolerance as assessed with the plasma glucose incremental area under the curve (iAUC) after an intravenous glucose tolerance test was not different between LDim and LD, either when the two rhythms in the LDim animals were aligned (P = 0.715) or misaligned (P = 0.368). Within the animals exposed to LDim, there was no difference in iAUC between the misaligned and the aligned state (P = 0.902) (Supplementary Figure 4).

Discussion
In the present paper we show that in the male Wistar rat, LDim strongly reduced the daily rhythms in sleep-wake behavior. LDim acutely increased sleep in the active phase during the first dim light (active) phase with a further increase over the following days, whereas LDim gradually decreased sleep during the light (inactive) phase. The reduced sleep-wake rhythm amplitude was paralleled by reduced rhythms in the 16-19 Hz EEG frequency domain and SCN clock gene expression. In a circadian analysis, LDim surprisingly induced an endogenous free running rhythm in locomotor activity with a period of ~25 h that interacted with the entrained 24 h rhythm. The reduced daily rhythms of sleep, food intake, and energy expenditure in LDim did not cause obesity or impaired glucose tolerance.
Already during the first 12 hours of dim light exposure, LDim reduced waking and increased NREM and REM sleep. This acute increase in NREM and REM sleep may be a direct effect of the dim light on sleep regulatory areas of the brain such as the ventrolateral preoptic nucleus or the ventral subparaventricular zone, since these areas receive direct information from intrinsically photosensitive retinal ganglion cells (ipRGCs) 27 and ipRGCs are sensitive to dim light levels 28 . Alternatively, the acute increase of NREM and REM sleep due to LDim may be mediated by decreased melatonin levels, as the dim light levels used are sufficient to decrease melatonin levels 29 and increased melatonin has been reported to reduce REM and NREM sleep in rats 30 . The effects of LDim on sleep increased over time. While LDim did not affect vigilance states in the light phase directly following the first 12 hours of dim light exposure, after 7 days LDim decreased NREM sleep during the light phase, and after 14 days LDim also increased waking during the light phase. The rhythm in NREM sleep slow wave activity, representing homeostatic sleep drive in the rat 31-33 was nearly abolished after 2 weeks of LDim exposure, in accordance with a near-equal distribution of sleep over the 24-h cycle. LDim also strongly decreased the rhythm in the 16-19 Hz frequency domain, which is within the NREM sleep EEG frequency range identified to reflect the rhythm of the endogenous circadian timing system in the Wistar rat 33,34 .
Together, the gradual increase of the effect of LDim on waking, NREM sleep, and REM sleep over the days, and the loss of rhythm in the 16-19 Hz frequency domain indicated an effect of LDim on the endogenous circadian timing system. To further investigate the effect of LDim on the endogenous period and rhythm strength, we exposed animals to LDim for 30 days followed by continuous darkness. Strikingly, in all animals LDim induced a free running rhythm in locomotor activity with a period of 25.1 h that interfered with the 24-h rhythm. LDim persistently affected the endogenous rhythm, since the transfer from LDim to DD resulted in a free running rhythm with a period of 24.8 h and low rhythmic strength, whereas the transfer from LD to DD resulted in a free running rhythm with a period that was close to 24 h and a higher rhythmic strength. Theoretically, the occurrence of two interacting rhythms in locomotor behavior in LDim can be due to either two separate endogenous rhythms or one free running endogenous rhythm with a period of 25.1 h in combination with one rhythm that is synchronized to the external 24 h rhythm (masking). After transfer from LDim to DD, rhythmic strength was lower in the initial 10 days (with three animals being virtually arrhythmic), compared to the second 10 days in DD, which supports the notion that the two rhythms may be two desynchronized endogenous rhythms that gradually resynchronize over time in DD.
In a control experiment we verified whether the occurrence of two interacting free running rhythms was specific for LDim and not due to continuous exposure to minimally 5 lux. Indeed, under continuous dim light (DimDim) or continuous light (LL), no rhythm splitting was observed. Instead, in DimDim, rats showed a single free running rhythm with a period duration of 24.9 hr, which was comparable to the period duration of the free running component under LDim. The increased free running period duration in continuous dim light compared to the free running period in continuous darkness after LD is in line with Aschoff 's rule that predicts an increase of period duration with increasing intensity of light exposure in nocturnal animals [35][36][37] . The complete loss of rhythmicity that we observed in LL has been reported before under high light intensities [37][38][39] .
The interaction within one animal of two rhythms in locomotor activity with a different period, also called internal desynchronization, has been described previously in various desynchronization paradigms in the vole 40 , the mouse 41,42 , and the rat 43 . However, to the best of our knowledge, two interacting rhythms in locomotor activity have never been described in a 12:12 L:D lighting schedule. Thus, our present data, for the first time, indicate that internal desynchronization can occur in a situation when the light-dark cycle adheres to the natural cycle duration of 24 hr. This finding may have translational consequences, since humans usually adhere to a 24-hr cycle.
The rodent and human circadian timing system share many common features. Both in rodents and humans, the circadian effects of light are mediated by ipRGCs. In both species, these ipRGCs detect light with the photopigment melanopsin that has maximal sensitivity for blue light with a wavelength of approximately 480 nm, and integrate this information with input from the classical photoreceptors in retinal rods and cones 44 . Subsequently, ipRGCs transfer the integrated light information to hypothalamic areas including the SCN 27,44 . The SCN rhythm of firing activity and neuropeptide expression is largely similar between nocturnal and diurnal mammalian species 45 . Despite these similarities, there are differences between the rodent and human circadian timing system in the sensitivity to light 44 : In humans for example, light with an intensity of 1 lux is barely sufficient to entrain the circadian pacemaker 46 , whereas in rodents, approximately 10-fold lower light levels clearly affect SCN neuronal firing 47 and circadian period 48 . In the present paper we used 5 lux as a light stimulus in order to replicate commonly used dim light conditions 18,24,25 . We are aware that lux is a measure defined by a weighting function specifically designed for the spectral sensitivity of the human image-forming visual system. Unfortunately, at present there is no validated alternative weighting function for the circadian effects of light 44 , although there have been some first attempts 49,50 .
The translational value of our findings is limited to the effects of dim light at night on the circadian timing system, since the direct effects of light at night is opposite between species: in rodents bright light induces sleep 16 whereas in humans bright light induces waking 51 .
With these caveats in mind, we believe that it is important to realize that internal desynchronization due to light at night as described in the present study, might be responsible for the association between light at night and insomnia in humans. It is interesting to note that the light intensity of 5 lux is frequently encountered in human bedrooms 13 , but keeping in mind the differences in sensitivity to light between the rodents and the human circadian timing system, it will be important to investigate if these light levels also cause internal desynchronization in diurnal species.
It is well known that the rodent SCN contains multiple cell groups that can function as separate pacemakers with a different period length in terms of clock gene expression 43,52 and neuropeptide secretion [53][54][55] . In Wistar rats subjected to a 22-h LD cycle, two interacting rhythms in behavior were coupled to two separate subdivisions of the SCN with the dorsomedial SCN representing the rhythm with a period of 24.9 h, and the ventrolateral SCN representing the rhythm with a period of 22 h 43 . In order to investigate the representation within the SCN of the two rhythmic components that we observed in locomotor behavior, we analyzed SCN clock gene expression at the time when the two rhythmic components in behavior were aligned or misaligned. LDim caused an overall reduction of the amplitude of the daily rhythm of Per1 and Arntl expression, but no significant desynchronization between ventrolateral and dorsomedial SCN could be detected. Possibly, the sensitivity of the two time-point analysis is insufficient to detect two independent rhythms in our model. Alternatively, the spatiotemporal organization of the two presumed pacemakers within the SCN may be less pronounced in our model compared to the 22-h day paradigm 43 . Future studies with transgenic clock-gene luciferase rats 56 may elucidate the anatomical representation of the free running rhythm.
Thus, we introduce a rat model for the effects of dim light at night on the circadian control of sleep. In contrast to the Swiss Webster mouse model 25 , we found profound disturbances of vigilance state distribution and sleep architecture. Based on the mouse model 18,24 , we expected that the circadian disruptions would have adverse metabolic consequences. However, despite increased daytime food intake during LDim on both chow and HFD, we did not observe effects of LDim on body weight, adiposity or glucose tolerance. On the chow diet, animals exposed to LDim did show a slightly smaller increase of food intake over the weeks compared to animals exposed to LD. Since body weight gain was not affected by the light schedule, this suggests that LDim may induce slightly more weight gain per g of food consumed (i.e. higher feeding efficiency). However, when the animals were induced to become obese with a high fat diet, this effect was not reproduced. The absence of metabolic effects contrasts to the Swiss Webster mouse model, where increased daytime food intake due to dim light at night causes obesity and decreased glucose tolerance 18,24 . The differences between the mouse model and our rat model may be attributable to species differences in the susceptibility to sleep disturbance or metabolic disease 57,58 or to the difference in light phase duration since the Swiss Webster mouse model utilized a 16:8 18 or 14:10 24,25 L:Dim cycle whereas we used a 12:12 L:Dim cycle.
In conclusion, we present the first rodent model for disrupted circadian control of sleep-wake rhythms due to light at night. In male Wistar rats, LDim strongly reduced the daily amplitude of sleep-wake and feeding-fasting behavior. This amplitude reduction was due to a direct sleep promoting effect combined with reduced SCN output strength due to internal desynchronization between the environmentally entrained 24 h rhythm and the LDim induced free running rhythm of ~25 h. We show that internal desynchronization is possible in a 12:12 L:D cycle, and therefore identified a potential mechanism for the association between light at night and insomnia in humans. Taken together, the detrimental effects of artificial light at night on the circadian control of sleep warrant attention from policy makers, researchers, and anyone with the control over a light switch.

Methods
Animals. Male Wistar WU rats (Charles River, Germany) were used for all experiments. Before the start of experimental procedures rats were given at least one week time for accommodation to the animal facility with Scientific RepoRts | 6:35662 | DOI: 10.1038/srep35662 standard chow and water available ad libitum. Experimental protocols were approved by the Animal Experiments Ethical Committees of the Leiden University Medical Center and/or the Royal Netherlands Academy of Arts and Sciences, and the study was performed in accordance with Dutch laws.
Light schedules. During accommodation, rats were exposed to a 12-h light (150-200lux):12-h dark (LD) schedule. To investigate the effect of dim light at night, rats were exposed to a 12 h light (150-200 lux):12 h dim white light (5 lux) (LDim) schedule, while control animals continued exposure to the LD schedule. During vigilance state and circadian activity experiments, dim light was emitted by white fluorescent tubes placed above the cage. During SCN clock gene expression and metabolic experiments, dim light was emitted by white LED lights placed around the cage. All light intensities were verified in the animal gaze direction with an AvaSpec 2048-SPU (Avantus BV, the Netherlands) light meter. The light source spectral power distributions are shown in Supplementary Figure 5. Vigilance state experiment. Rats (n = 13; bodyweight 283 ± 5 g) were anesthetized with Ketamine/ Xylazine/Atropine, fixed in a stereotaxic frame and implanted with 2 electroencephalogram (EEG) screw electrodes (Plastics One, Roanoke, VA, USA; 1.4 mm diameter) and 2 electromyogram (EMG) patch electrodes (Plastics One; 0.35 mm diameter) as described previously 26,[31][32][33] . One EEG electrode was placed over the right hemisphere above the sensory cortex, while the other was placed above the cerebellum. The EMG electrodes were inserted between the neck muscles and the skin. The wires of the electrodes were set in a plastic pedestal (Plastics One) which was fixed to the skull with dental cement and 3 additional support screws.
After 7-10 days recovery from surgery in individual housing, rats were divided into two bodyweight-matched groups and exposed to either LDim (n = 8) or LD (n = 5). Twenty-four h EEG/EMG recordings were performed at baseline (day 0, prior to the LDim schedule), day 1, day 7 and day 14. Recordings on day 0 and day 1 started at ZT12, recordings on day 7 and day 14 started at ZT0. The EEG and EMG were recorded with a portable system (Institute of Pharmacology and Toxicology, University of Zurich, Zurich, Switzerland) as described previously 59,60 . Before each recording, a calibration signal (10 Hz sine wave, 300 μ V peak-to-peak) was recorded on the EEG and EMG channel. Both signals were amplified (amplification factor ~2000), conditioned by analogue filters (high-pass filter − 3 dB at 0.16 Hz) and sampled at 512 Hz. The signals were filtered through a digital Finite Impulse Response filter: EEG low-pass filter at 30 Hz and EMG band-pass filter between 20 and 40 Hz.
The vigilance states (waking, NREM sleep and REM sleep) were scored off-line in 4-s epochs by visual inspection of the raw EEG and EMG signal, as well as EEG power density in the slow-wave range (0.75-4.0 Hz) according to standardized criteria for rats 26,61  Circadian experiments. Rats (n = 8, bodyweight 254 ± 1 g) were individually housed in transparent plastic cages and locomotor activity was monitored by passive infrared detectors using Actimetrics software (Wilmette, Il, USA). In circadian experiment A, animals were first exposed to 10 days LD and subsequently to 30 days LDim, 20 days constant darkness (DD), re-entrainment during 30 days LD and finally 10 days DD. In circadian experiment B we investigated the effect of continuous light intensity on circadian activity by exposing the rats (n = 8, bodyweight 251 ± 2 g) consecutively to at least 10 days LD, 16  SCN clock gene expression experiment. Individually housed rats (n = 42, bodyweight 254 ± 4 g) were exposed to either LDim or LD and locomotor activity was measured using pressure-sensitive baseplates 5 . For LDim animals, the moments were determined when the activity phase of the free running rhythm and the activity phase of the 24-h rhythm were either aligned or misaligned, as determined by visual inspection of the actogram combined with F-periodogram analysis 63,64 . Rats were divided into 6 groups (n = 7 per group) and sacrificed accordingly: 1) LDim aligned ZT6, 2) LDim aligned ZT18, 3) LDim misaligned ZT6, 4) LDim misaligned ZT18, 5) LD ZT6 and 6) LD ZT18. Animals were briefly anesthetized with 80% CO 2 and decapitated. Brains were frozen on dry ice and stored at − 80 °C. Coronal 20 μ m slices of the SCN were obtained with a cryostat. In situ hybridization of Per1 and Arntl (Bmal1) were performed using riboprobes of rPer1 (kindly provided by Dr. H. Okamura, Kyoto University, Japan) and mArntl (mBmal1) 65 genes. Antisense RNA probes were generated with an in vitro transcription kit (Maxiscript; Ambion, Austin, TX, USA). Hybridization was carried out as described previously 66 . Slices and radioactive standards were exposed to an autoradiographic film (Biomax MS-1 Kodak, Sigma-Aldrich, St Louis, MO, USA). Quantitative analysis of the autoradiograms was performed using the ImageJ software (W. Rasband, National Institutes of Health, Bethesda, MD, USA). A Nissl staining was performed on the same slides and the section containing the mid SCN was selected. The optical density of the whole SCN, the ventromedial SCN and the dorsolateral SCN were measured by a researcher (JMen) who was blind for treatment, Relative optical density was determined by subtracting the background intensity (measured in the anterior hypothalamic area) from the signal for each animal. Left and right sided relative optical densities were averaged and the dorsomedial/ ventrolateral ratio 43  Individually housed rats were divided into two bodyweight-matched groups and exposed to either LDim or LD. For the assessment of food intake and energy expenditure, rats were placed in PhenoMaster/LabMaster calorimetric cages (TSE Systems GmbH, Bad Homburg, Germany) in week 0 (baseline, before start of dim light exposure), and 1, 2 and 7 weeks after the start of dim light exposure. Rats were sacrificed after 8 weeks. The second metabolic experiment was performed with 16 rats (bodyweight 196 ± 1 g) on a high fat diet (HFD) (D12451 high fat diet, Research Diets, Inc, New Brunswick, New Jersey, USA). Again, rats were divided into two bodyweight-matched groups and exposed to either LDim or LD. HFD animals were placed in the calorimetric cages in week 0 (baseline, before start of dim light exposure), and 1, 2, 3, 4, 5, 8 and 9 weeks after the start of dim light exposure, and HFD animals were sacrificed after 11 weeks. Rats were given at least 2 hours to accommodate to the calorimetric cages before the 48-hr recordings started.
After sacrificing the animals, the mesenteric, epidydimal, perirenal and subcutaneous WAT depots, adrenals and thymus from the right side of the rat were collected and weighed.
Glucose tolerance experiment. Individually housed rats were divided into two bodyweight-matched groups (n = 7 per group) and exposed to either LDim or LD for at least 7 days before surgery. Locomotor activity was measured using pressure-sensitive baseplates 5 . The rats (bodyweight 320 ± 11 g at surgery) were anesthetized with fentanyl/fluanisone/midazolam, a silicone catheter was inserted into the right atrium via the jugular vein and attached to the skull with 4 screws and dental cement 5 . Rats were given at least 1 week to recover, and glucose tolerance was assessed at ZT6 twice in every rat: at the time when the activity phases of the free running and 24 h rhythms were aligned and when the activity phases were misaligned (as determined by visual inspection of the actogram combined with F-periodogram analysis 63,64 ). One day before the intravenous glucose tolerance test (IVGTT) 5 rats were attached to a head stage that allows infusion and blood sampling in freely moving animals. On the day of the experiment, food was removed 4 h before the glucose injection. A 1000 mg/kg bodyweight glucose bolus (Sigma-Aldrich) was injected at t = 0 and blood samples were obtained from the catheter at t = − 1, 5, 10, 20, 30 and 60 min. Glucose was measured immediately with a FreeStyle Freedom Lite glucose measurement device (Abott Diabetes Care, Alameda, CA, USA).

Statistical analysis.
Normally distributed variables are expressed as mean and standard error of the mean (SEM). Vigilance state and EEG data were analyzed with repeated measures ANOVA with the factors Phase (Light, Dark/Dim), Day (0, 1,7,14) and their Interaction. Power density spectra of NREM sleep were analyzed with ANOVA with the factors Frequency (0.25-25.0 Hz), Day (0, 1,7,14) and their Interaction. For circadian locomotor activity data, the rhythm period and strength were determined with F-periodogram analysis 63 as previously described 64 . Rhythm strengths were analyzed by ANOVA with the factor Light schedule (LD, LDim, DD). SCN gene expression data were analyzed with ANOVA using the factors Time (ZT6, ZT18), Group (LD, LDim-aligned, LDim-misaligned) and their Interaction.
Data from the calorimetric cages were analyzed with a linear mixed model in order to correctly incorporate missing data, with Light schedule (LD, LDim), Week (week 0-9) and their Interaction as fixed effects. The covariance structure 'Compound Symmetry' was selected based on the Akaike's Information Criterion (AIC).
If a significant main effect was detected in ANOVA or linear mixed model analyses, post hoc tests were performed with an independent samples two-tailed student's t-test (for 2 groups with independent samples), a related samples two-tailed student's t-test (for 2 groups with related samples) or Duncan's test (for > 2 groups).
For glucose tolerance data, the iAUC was calculated for individual animals as described previously 67 . The effect of LDim exposure on body weight, organ weight and glucose tolerance was assessed with a student's independent samples two-tailed t-test or a student's related samples two-tailed t-test where appropriate.
All statistical analysis were performed with SPSS statistics (v21, SPSS, Inc) using a significance level of 0.05.