Dim evening light shifts daily body rhythms, sleep, and short-term memory together

To examine the effects of 20-lux DLE during Zeitgeber times (ZT)12 to 16 on circadian rhythms, locomotor activity phases (onset, midpoint, and offset) were determined from passive infrared sensor (PIR) recording across each 24-h cycle. Among the three phase markers, activity offset is known to show the highest variability (40); this is in agreement with our data under LD. Under DLE, C57BL/6 wild-type (WT) mice exhibited changes in alignment of all three parameters with respect to the light phase (Fig. 1A). More specifically, activity onsets were delayed by 3.59 ± 0.46 h [main effects of Lighting F(1, 10) = 64.31, P < 0.001], whereas activity midpoints and offsets were delayed by 2.57 ± 0.37 and 2.52 ± 0.83 h [F(1, 10) = 43.94, P < 0.001 and F(1, 10) = 10.64, P = 0.009, respectively]. Changes in activity midpoints and offsets, which occurred hours after the DLE, suggest that these are not the acute effect of activity suppression by DLE. Furthermore, under constant darkness (DD), activity rhythms in mice with prior exposure to DLE free ran at later clock times than mice with prior exposure to LD (n = 8 Albumin-Cre;Dbp^KI/+ mice per group as described in Molecular Rhythms Are Delayed under DLE), with an overall phase difference of ∼4 h [main effects of Prior Lighting: F(1, 14) = 6.77, P = 0.021; F(1, 14) = 6.30, P = 0.025; and F(1, 14) = 6.20, P = 0.026, for onsets, midpoints, and offsets, respectively]. The persistence of DLE-induced phase delay under DD confirmed that these effects were not the result of the suppression of locomotor activity by light.

As the oestrous cycle in female mice (which lasts for 4 to 5 d) has been shown to affect wheel-running activity rhythms (41), we assessed the possibility that the effects of DLE on PIR activity rhythms could vary between sexes. However, there was no Lighting × Sex interaction, indicating that DLE exerted comparable effects in both sexes [Lighting × Sex interactions F(1, 10) = 1.66, P = 0.23; F(1, 10) = 0.06, P = 0.81; and F(1, 10) = 2.68, P = 0.13, for onsets, midpoints, and offsets, respectively].

Similar DLE-induced phase delays were observed in OPN4-deficient (Opn4^−/−) mice of both sexes. Under DLE, activity onsets, midpoints, and offsets of Opn4^−/− mice were delayed by 2.74 ± 0.18, 2.32 ± 0.95, and 1.61 ± 0.61 h [main effects of Lighting F(1, 10) = 215.25, P < 0.001; F(1, 10) = 87.14, P < 0.001; and F(1, 10) = 7.53, P = 0.021, respectively; Fig. 1B]. Although the size of phase shifts appeared smaller in Opn4^−/− mice, comparisons between WT controls and Opn4^−/− mice indicated that DLE effects were statistically indistinguishable between genotypes (main effects of Genotype Ps > 0.08; Lighting × Genotype interactions Ps > 0.09). Thus, like other nonvisual responses to light (42–44), the conventional retinal circuitry can support DLE-induced phase delays in the absence of OPN4. Under DLE, the power of ∼24 h rhythms persisted in both WT and Opn4^−/− mice; however, there was an increase in oscillatory power at ∼8 h in both groups as revealed by continuous wavelet transform (45) (SI Appendix, Fig. S1↗). This is likely to be an artifact of the shortened, 8-h night under DLE rather than reflecting an increase in ultradian rhythms per se.

To examine changes in sleep patterns, for each 10-s bin of the PIR recording, the mouse’s behavioral state was assigned as either awake (0) or asleep (1), where “1” was defined as sustained PIR inactivity for at least four uninterrupted 10-s bins; if this was not the case, a value of “0” was assigned (Fig. 2A). Sleep proportion was calculated as the number of bins assigned with a value of “1” divided by the total number of 10-s bins during that time window; the duration of a sleep bout was defined as the number of uninterrupted 10-s bins assigned with a value of “1.” It has been validated that ≥40 s immobility is highly correlated with electroencephalogram (EEG)-defined sleep times across the 24-h day, with Pearson’s r ≥ 0.94 (46–48).

Total immobility-defined sleep times were unaffected under DLE (SI Appendix, Table S1↗). However, there were changes in the alignment of sleep patterns under DLE, both in terms of sleep proportion [Lighting × Time of Day interaction in WT: Greenhouse–Geisser corrected F_Ɛ(2,29) = 26.03, P < 0.001; Lighting × Time of Day interaction in Opn4^−/−: F_Ɛ(3,33) = 24.17, P < 0.001; Fig. 2B] as well as duration of sleep bouts [Lighting × Time of Day interaction in WT: F_Ɛ(1,14) = 6.40; P = 0.017; Lighting × Time of Day interaction in Opn4^−/−: F_Ɛ(1,19) = 14.10, P < 0.001; Fig. 2C]. Crucially, under DLE, sleep proportion was reduced and sleep bouts were shortened in the first 4 h of the light phase (simple effects of Lighting from ZT0 to 4 Ps < 0.005). Thus, DLE promoted sleep from ZT12 to 16 and postponed the accumulation of sleep pressure in the next morning. Comparisons between WT and Opn4^−/− mice did not reveal any genotype difference (main effects of Genotype Ps > 0.08; Lighting × Genotype interactions Ps > 0.1; Lighting × Time of Day × Genotype interactions Ps > 0.2).

As sleep and inactivity are often accompanied by body cooling (49), we examined if there was any change in skin temperature (T_skin) using infrared thermography (SI Appendix, Fig. S2A↗) as described in our previous study (50). T_skin provides an estimate of changes in core body temperature over time (50). The phase of T_skin offset was delayed under DLE (SI Appendix, Fig. S2B↗), corresponding to the delayed PIR activity rhythm and sleep profile. In addition, there was a 2% drop in T_skin (∼0.65 °C) during the 4-h DLE period as well as a 0.5 to 1% reduction in T_skin (∼0.15 to 0.35 °C) in the light phase (SI Appendix, Fig. S2C↗).

When the intensity of the 4-h evening light is the same as the prevailing light phase, this would become a long-day photoperiod. We compared locomotor activity rhythms and sleep under 16-h day/8-h night cycles (16:8 LD) versus DLE. Under 16:8 LD, activity onsets and midpoints were further delayed by 2.21 ± 0.49 and 0.99 ± 0.18 h relative to DLE [main effects of Lighting F(1, 22) = 20.80, P < 0.001 and F(1, 22) = 29.81, P < 0.001, respectively]. The difference in activity offsets under 16:8 LD versus DLE, 0.64 ± 0.36 h, was not significant (P > 0.09). Comparisons between WT and Opn4^−/− mice did not reveal any genotype difference in phasing (main effects of Genotype Ps > 0.4; Lighting × Genotype interactions Ps > 0.2) (SI Appendix, Fig. S3A↗). Similarly, effects of 16:8 LD on sleep were stronger relative to DLE [Lighting × Time of Day interactions F_Ɛ(2,49) = 26.74, P < 0.001 and F_Ɛ(1,31) = 3.98, P = 0.041 for sleep proportion and duration of sleep bouts, respectively]. Under 16:8 LD, mice slept more and exhibited longer sleep bouts during ZT8 to 16 and ZT20 to 24, but they slept less and showed shorter sleep bouts during ZT0 to 4 (SI Appendix, Fig. S3 B and C↗). There was no significant difference in sleep between WT and Opn4^−/− mice (main effects of Genotype Ps > 0.1; Lighting × Genotype interactions Ps > 0.5; Lighting × Time of Day × Genotype interactions Ps > 0.5). Thus, DLE exerts weaker effects on locomotor activity rhythms and sleep than a long day.

Misalignment of circadian rhythms is accompanied by changes in clock gene expression throughout the body (51). To investigate this possibility under DLE, we examined clock gene expression (Per2, Bmal1, Rev-erbα, Cry1, and Dbp) in the heart, liver, adrenal gland, and dorsal hippocampus at ZT2, ZT8, ZT14, and ZT20. Expression levels of target genes were normalized to the geometric mean of two housekeeping genes, Tbp and Gapdh (52); validation of these housekeeping genes is reported in SI Appendix, Supplementary Methods↗. Under LD, the phasing of clock gene expression observed was broadly consistent with patterns from previous mouse studies: Per2 peaked at early night, Bmal1 peaked near dawn, Rev-erbα peaked during the day, and Dbp peaked before the start of the night (51). Some of these molecular rhythms were delayed under DLE (Fig. 3A). Peaks of normalized Bmal1 and Cry1 messenger RNA (mRNA) levels in the adrenal gland were delayed from the dark phase to ZT2 [Lighting × Time of Day interactions F(3, 21) = 3.37, P = 0.038 and F(3, 21) = 3.69, P = 0.028, respectively]. In the dorsal hippocampus, the peak of normalized Per2 expression was shifted from ZT14 to ZT8 [Lighting × Time of Day interaction F(3, 21) = 3.17, P = 0.046]. In addition, centers of gravity (CoG)—providing an estimate of the acrophase (φ) of the molecular rhythm (53–55)—were consistently delayed under DLE; phase shifts (∆φ) were expressed as the difference in CoG between DLE and LD (SI Appendix, Fig. S4↗). When pooled across peripheral tissues, one-sample Student’s t tests (two-tailed) showed that mean ∆φ values of Rev-erbα (2.07 ± 0.57 h) and Dbp (1.81 ± 0.51 h) were different from the value of 0 [one-sample t(3) = 3.67, P = 0.035 and t(3) = 3.55, P = 0.038, respectively].

The expression of six hepatic genes involved in adipogenesis and lipid/glucose metabolism (Hmgcr, Hdac3, Ccrn4l, Pparγ, Npc1, and Cyp7a1) was also examined. Notably, DLE delayed the circadian phase of the cholesterol synthesis gene Hmgcr: it peaked at ZT8 under LD (SI Appendix, Fig. S5A↗) as reported in a previous study (56); however, this was delayed to ZT14 under DLE [Lighting × Time of Day interaction F(3, 22) = 5.037, P = 0.008]. The mean ∆φ value of all six hepatic genes under investigation was 2.35 ± 0.95 h [one-sample t(5) = 2.49, P = 0.055 (two-tailed); SI Appendix, Fig. S5B↗].

To provide an additional functional measure of hepatic circadian output, we recorded bioluminescence signals from freely moving Albumin-Cre;Dbp^KI/+ liver reporter mice, which expressed firefly luciferase in hepatocytes under the control of Dbp (57). Animals were kept either under LD or DLE for 1 wk; synthetic luciferin (CycLuc1) was then administered in drinking water, allowing hepatic bioluminescence emission to be detected and recorded in DD (Fig. 3B). Under DD, the strength of Dbp bioluminescence rhythm (as indicated by signal-to-noise ratios) and period length were indistinguishable between groups [main effects of Prior Lighting F(1, 14) = 1.84, P = 0.20 and F(1, 14) = 2.11, P = 0.17, respectively]. However, the acrophase of Dbp bioluminescence rhythm was delayed by an average of 2.34 h in mice with prior DLE exposure [main effect of Prior Lighting F(1, 14) = 37.13, P < 0.001; Fig. 3B]. Despite the shift in the hepatic circadian clock and metabolic rhythm, 2 wk of DLE did not affect body weight in C57BL/6 mice (SI Appendix, Fig. S5 C and D↗).

Given the DLE-induced delay in circadian physiology of ∼2 h, we examined if there was any consequence for short-term memory processes at 2 h into the light phase (ZT2) and 2 h after the light phase (ZT14). We used the spontaneous recognition memory task (Fig. 4A) (58), which is sensitive to effects of aberrant lighting (23, 59–61). Initial analyses confirmed that under LD, there was a day/night difference in short-term object and odor recognition memory performance, with better performance at ZT2 than at ZT14 [main effect of Time of Day F(1, 15) = 6.59, P = 0.021; SI Appendix, Fig. S6A↗]; this is consistent with our previous findings (60). Repeated recognition assessment 2 wk later did not alter the size or direction of this effect [main effect of Repeated Testing F(1, 15) = 0.038, P = 0.85; Time of Day × Repeated Testing interaction F(1, 15) = 0.60, P = 0.45], demonstrating the stability of this behavioral rhythm under LD.

To examine the effect of DLE, another group of mice first received recognition trials at ZT2 and ZT14 under LD and at the same time points following 2 wk of DLE exposure (Fig. 4A). In these mice, the diurnal pattern of recognition memory performance was reversed under DLE [Lighting × Time of Day interaction F(1, 10) = 9.88, P = 0.01; Lighting × Time of Day × Type of Stimulus interaction F(1, 10) = 0.13, P = 0.73; Fig. 4B]. Under LD, performance was better at ZT2 than at ZT14 [simple effect of Time of Day F(1, 10) = 6.21, P = 0.032]. By contrast, under DLE performance was better at ZT14 than at ZT2 [simple effect of Time of Day F(1, 10) = 4.96, P = 0.05]. This resulted in enhanced performance at ZT14 under DLE [simple effect of Lighting F(1, 10) = 8.67, P = 0.015; Fig. 4B] but not at ZT2 [F(1, 10) = 2.45, P = 0.15]. The improved test performance at ZT14 was due to reduced familiar object exploration [SI Appendix, Fig. S6 B↗, Right]. In the sample phase, there was a suggestion of elevated exploratory activity at ZT2 but reduced exploration at ZT14 under DLE (SI Appendix, Fig. S6 B↗, Left). However, these effects did not reach significance [Lighting × Time of Day interaction F(1, 10) = 4.81, P = 0.053; all simple effects Ps ≥ 0.057] and did not relate to the pattern of exploratory activity in the test phase (SI Appendix, Fig. S6B↗).

We then explored if the reversal of memory performance was related to changes in sleep history under DLE. To address this question, we determined the amount of sleep from time windows of varying widths (from 10 min up to 3 h in 10-min increments) preceding each recognition trial. Adopting a wider time window increased the goodness of the linear fit, R², between prior sleep and performance, and R² reached an asymptotic value of ∼0.9 at t = ∼2 h (Fig. 4 C, Left, data point in magenta). Notably, there was a strong linear relationship between preceding 2-h sleep and performance at the group level (Fig. 4 C, Right). When all individual cases were considered (11 mice × 8 trials per mouse), linear mixed-effects models (62) confirmed the benefits of preceding 2-h sleep on subsequent performance (likelihood ratio χ² = 7.73, P = 0.005). The bootstrap 95% CI of the effect of sleep was [+0.12, +0.62], which did not overlap with zero. By contrast, sleep history did not bear any relationship to sample exploratory activity (likelihood ratio χ² = 1.30, P = 0.25), which in turn was unrelated to test performance (likelihood ratio χ² = 0.33, P = 0.57); 95% CIs of these effects, [−58.73, +15.02] and [−0.002, +0.001], respectively, overlapped with zero. Together, our data suggest that prolonged wakefulness at night is associated with a decline, whereas DLE-induced sleep is associated with a facilitation in short-term memory process.

To assess changes in brain states under DLE, we quantified brain immunofluorescence cFos levels in naïve mice at ZT2 and ZT14. This provides a molecular correlate of neuronal activity within the ∼1 to 2 h time window prior to learning in the recognition memory task. We found that the SCN and medial prefrontal cortex (mPFC) were differentially modified under DLE as indicated by the Lighting × Time of Day × Region interaction [F(1, 8) = 60.23, P < 0.001]. Under LD, there were more cFos+ cells in the SCN at ZT2 than at ZT14 [simple effect of Time of Day F(1, 8) = 39.23, P < 0.001], but an opposite, nocturnal pattern was observed in the mPFC [simple effect of Time of Day F(1, 8) = 13.74, P = 0.006], indicating that SCN and mPFC were out of phase under LD. Importantly, the diurnal pattern in the SCN under LD was attenuated but remained significant under DLE [simple effect of Time of Day F(1, 8) = 7.92, P = 0.023] due to increases in cFos+ cell count at ZT2 and ZT14 [simple effects of Lighting F(1, 8) = 9.33, P = 0.016 and F(1, 8) = 42.30, P < 0.001, respectively; Fig. 5A]. Moreover, patterns of cFos signals in the SCN dorsal and ventral subregions were dissociated under DLE (SI Appendix, Fig. S7↗), which may be suggestive of reorganization in the SCN (17–19). By contrast, the nocturnal pattern in the mPFC under LD was reversed by DLE [simple effect of Time of Day F(1, 8) = 75.76, P < 0.001]: there was an increase in the number of cFos+ cells at ZT2 but reduced signals at ZT14 [simple effects of Lighting F(1, 8) = 65.32, P < 0.001 and F(1, 8) = 18.74, P = 0.003, respectively; Fig. 5B]. The reversal of cFos signals was found in both superficial (layers one to four) and deep layers (layers five/six) of the mPFC (SI Appendix, Fig. S8↗).

The preoptic hypothalamus (POA) is known to regulate body cooling and promote sleep (63–67), whereas the lateral and dorsomedial hypothalamus (LHA and DMH) is known to promote wakefulness (68–70). Indeed, cFos signals in these hypothalamic regions were differentially modified by DLE as indicated by the Lighting × Time of Day × Region interaction [F(2, 16) = 12.47, P = 0.001]. In the POA, there was an increase in the number of cFos+ cells at ZT14 (SI Appendix, Fig. S9↗). By contrast, in the LHA, there was a decrease in cFos+ cell count at ZT14 (SI Appendix, Fig. S10 A and B↗). In addition, cFos+ cell counts in the LHA and DMH were positively related to each other and to the cortex but not to the POA (SI Appendix, Fig. S10C↗). Thus, cFos signals in the LHA/DMH versus POA were modified in a way that corresponds to their established roles in regulating wakefulness and sleep, respectively.

Exposure to long-day photoperiods for 1 to 2 wk is known to decrease the number of dopamine tyrosine-hydroxylase–expressing (TH+) cells in the paraventricular nuclei of the hypothalamus (PVN), increasing behavioral immobility and reducing exploratory activity in rats and mice (20, 21). Similarly, seasonal variation in photoperiod is associated with changes in TH levels in the human midbrain (22). To assess this possibility under DLE, we examined immunofluorescence TH levels in the PVN as well as in the ventral tegmental area (VTA) and substantia nigra pars compacta (SNc), which provide major dopaminergic projections to different parts of the brain. DLE did not reduce the number of TH+ cells in the PVN [main effect of Lighting F(1, 8) = 0.51, P = 0.50; Lighting × Time of Day interaction F(1, 8) = 1.08, P = 0.33; SI Appendix, Fig. S11↗, Upper] or in the VTA/SNc [main effect of Lighting F(1, 8) = 0.82, P = 0.39; Lighting × Time of Day interaction F(1, 8) = 0.24, P = 0.64; SI Appendix, Fig. S11↗, Lower].

A total of 111 mice were used in this study. In experiments examining home cage activity and sleep, there were 12 WT mice of C57BL/6 background (Opn4^+/+) and 12 Opn4^−/− mice (equal numbers of ♀ and ♂) obtained from crossbreeding heterozygous Opn4^+/− mice (104). Detailed analyses of locomotor activity and immobility-defined sleep under normal and abnormal lighting conditions did not reveal any difference between sexes or genotypes; thus, male C57BL/6 WT mice (Envigo; RRID: IMSR_JAX:000664↗) were used in subsequent experiments. In the DD experiment in which in vivo hepatic bioluminescence and locomotor activity were examined, 16 Albumin-Cre;Dbp^KI/+ liver reporter mice (9♀ and 7♂) were used (57). All mice were singly housed with ad libitum access to food and water. Cages were placed within light-tight ventilated chambers (LTC) equipped with multiple cool-white light-emitting diodes (LEDs) (Luxeon Star LEDs; Quadica Developments), providing a light level of 200 lx during the day; the spectral power distribution of the LEDs consisted of a higher and narrower peak at ∼460 nm and a lower and broader peak at ∼560 nm. Each LTC was also equipped with PIR for long-term recording of locomotor activity (49). Animals were kept in LTC under 12-h day/12-h night cycles (12:12 LD) for at least 2 wk, before receiving dim light (20 lx) in the evening from ZT12 to 16 (DLE) or 16-h day/8-h night cycles (16:8 LD). The temperature of the animal holding room was maintained at 19 to 21 °C. Experimental procedures involving C57BL/6 WT and Opn4^−/− mice were conducted at the University of Oxford, England in accordance with the United Kingdom Animals (Scientific Procedures) Act 1986 under Project License PE4ED9D2C/PP0911346 and Personal Licenses I869292DB and IDB24291F. The DD experiment involving Albumin-Cre;Dbp^KI/+ mice was conducted at Smith College, Massachusetts in accordance with the standard of the Association for Assessment and Accreditation of Laboratory Animal Care International.

Raw PIR data with a temporal resolution of 10 s were first smoothed with a 1-h moving average across the entire recording period. Following ref. 105, activity onsets and offsets were defined as the intersections between a short moving average (3-h window) and a long moving average (24-h window) of the smoothed time series. More specifically, for each time bin (i), the difference between short and long moving averages (Δ_i) was determined, and the product of Δ_i−1 and Δ_i was calculated. The daily onset of activity was the time point at which the short moving average first exceeded the long moving average, that is, when Δ_i−1Δ_i < 0 and Δ_i−1 < 0, whereas daily activity offsets were the time points at which the short moving average dropped below the long moving average, that is, when Δ_i−1Δ_i < 0 and Δ_i−1 > 0. Activity midpoints were defined as the time points at which total activity in the preceding 8 h was the same as total activity in the subsequent 8 h (105). To implement this, we determined the difference between preceding 8-h activity and subsequent 8-h activity (Δ_i) for each time bin and calculated the product of Δ_i−1 and Δ_i. Activity midpoints were estimated as the time points at which Δ_i−1Δ_i < 0 and Δ_i−1 < 0. Activity onset, midpoint, and offset calculations were conducted in R (RRID: SCR_001905↗).

A total of 16 Albumin-Cre;Dbp^KI/+ mice expressing firefly luciferase in hepatocytes under the control of Dbp were used to measure hepatic Dbp bioluminescence signals in vivo (57). They were first singly housed under LD for at least 8 wk. Half of them were exposed to DLE for a week, whereas the other half remained under LD. Synthetic luciferin CycLuc1 (Tocris Bioscience) was then administered by dissolving CycLuc1 in dimethyl sulfoxide (DMSO) and diluting the 50 mM stock solution to 0.1 mM in drinking water. The animal’s back and abdominal hair was shaved to optimize detection of hepatic bioluminescence under DD. The mouse cage was placed inside the dark LumiCycle In Vivo recording unit (Actimetrics) equipped with two photomultiplier tubes above the cage for detecting bioluminescence emission. A programmable shutter covered the photomultiplier tubes periodically (1 min in every 15 min) to record background counts and obtain background-corrected bioluminescence counts every minute. Hepatic bioluminescence was recorded in DD for 5 d. Discrete wavelet transform was conducted on bioluminescence data as described previously (107, 108). Locomotor activity under DD was recorded from PIR (K-940; Visonic); activity onset, midpoint, and offset were determined as described in Locomotor Activity Phase Markers.

Recognition testing took place in a 25 × 25 × 25 cm transparent acrylic arena with two wallpapers attached to the exterior walls. One wallpaper was a chequerboard pattern with 4 × 4 cm black and white squares. The wallpaper on the opposite side was black with a white symmetrical five-point star shape of 14 × 14 cm. An infrared camera and a white LED light were positioned above the center of the arena. The light level was 50 lx on object trials and 0 lx on odor trials. SI Appendix, Table S3↗ provides a description of all eight different objects and eight odors used in the experiment. There were multiple replicates of each object, and different replicates of the same object were presented in sample and test phases. The arena and objects were wiped with 70% ethanol after each phase.

Mice received 10-min acclimatization trials in the empty arena at ZT8 (midway between ZT2 and ZT14) for 5 d. They then received object and odor recognition trials at ZT2 and ZT14 and at the same ZTs after 2 wk of exposure to DLE. During the sample phase of each trial, two identical replicates of an object or an odor cue were placed in the arena; the sample phase was terminated after 10 min. After a 3-min delay, novelty preference was assessed in the arena, now containing a third replicate of the familiar sample and a novel stimulus; the test phase was terminated after 3 min. For six of the animals (subgroup 1), the order of recognition trials under LD was the following: ZT2 (odor), ZT14 (odor), ZT14 (object), and ZT2 (object); whereas for the remaining animals (subgroup 2), the arrangement was reversed: ZT14 (odor), ZT2 (odor), ZT2 (object), and ZT14 (object). Under DLE, the order of assessment was ZT2 (object), ZT14 (object), ZT2 (odor), and ZT14 (odor) for subgroup 1; for subgroup 2, the order was ZT14 (object), ZT2 (object), ZT14 (odor), and ZT2 (odor). The order of recognition assessment is summarized in Fig. 4A. The identity of novel and familiar stimuli and their spatial positions in the arena at test were counterbalanced in order to take into account any potential stimulus-specific or location-specific bias. Automated tracking of exploratory activity was conducted in ANY-maze (Stoelting; RRID:SCR_014289↗), which tracked the position of the mouse's snout every 0.1 s and recorded stimulus exploration times. Test performance was expressed as the proportion of time spent with the novel stimulus. Rodents show the strongest novelty preference within the first 30 s of stimulus exploration (95). This declines rapidly over time and reaches chance level by the second and third minutes of testing (109, 110). Thus, our main analyses focused on 0 to 60 s of test performance; data from 60 to 120 s and 120 to 180 s of the test phase are summarized in SI Appendix, Table S4↗. As sample exploratory activity levels were >25 s across the four lighting/ZT conditions (range = 25.25 to 130.60 s when pooled across object and odor trials), with average exploration times (SI Appendix, Fig. S6 B↗, Left) similar to previous mouse studies (111, 112), no post hoc or arbitrarily defined threshold was used to exclude any data point from analyses.

Mice housed under LD or DLE were injected intraperitoneally with sodium pentobarbitone at ZT2 or ZT14. Immediately after the loss of the pedal reflex, they were perfused transcardially with phosphate-buffered saline (PBS) and 4% paraformaldehyde (PFA). The brain was extracted from the skull, kept in 4% PFA overnight, rinsed with PBS the next day, and transferred into 30% sucrose solution for cryopreservation. After 2 to 3 d, brains were sectioned coronally at 50 μm on a freezing microtome (Leica). Based on Franklin and Paxinos’ atlas (113), coronal sections near Bregma +1.98, +0.14, −0.82, and −1.94 mm were selected to examine cFos signals in the mPFC, POA, SCN, and LHA/DMH, respectively; these correspond to plates 100048576_117, 100048576_209, 100048576_241, and 100048576_281 retrieved from the Allen Mouse Brain Atlas (114). The selected sections for each region from all four conditions were processed on the same days to minimize variability in staining quality. Sections were first washed in PBS, blocked in normal donkey serum (Jackson ImmunoResearch; RRID: AB_2337258↗) for at least 90 min, and incubated at 4 °C with rabbit anti-cFos (dilution 1:4,000; Synaptic Systems; RRID: AB_2619946↗), a molecular correlate of recent neuronal activity. In addition, for mPFC sections, the rat anti-Ctip2 antibody (dilution 1:250; Abcam; RRID: AB_2064130↗) was added to serve as a marker for deep cortical layers (layers five/six); for SCN sections, the mouse anti-GAD67 antibody (dilution 1:500; Merck Millipore; RRID: AB_2278725↗) was added to serve as a marker for the SCN; and for other hypothalamic sections, the chicken anti-TH antibody (dilution 1:500; Abcam; RRID: AB_1524535↗) was added to mark hypothalamic TH-immunoreactive cell bodies and fibers. After 72 h of primary antibody incubation, POA, SCN, and LHA/DMH sections were washed in PBS and incubated at room temperature with Alexa Fluor 488 donkey anti-chicken, Cy3 donkey anti-rabbit, and Alexa Fluor 647 donkey anti-mouse secondary antibodies (dilution 1:500; Jackson ImmunoResearch; RRID: AB_2340375↗, RRID: AB_2307443↗, and RRID:AB_2340863↗, respectively), whereas mPFC sections were incubated with Alexa Fluor 488 donkey anti-rabbit and Cy3 donkey anti-rat secondary antibodies (Jackson ImmunoResearch; RRID: AB_2313584↗ and RRID: AB_2340667↗, respectively). After 4 h of secondary antibody incubation, sections were washed in PBS. Finally, stained sections were mounted on glass slides and coverslipped with Vectashield (Vector Laboratories; RRID: AB_2336789↗). Processed sections were scanned using a confocal microscope (Fluoview FV1000; Olympus) under 10× and 20× objective lenses. Identical settings were used to acquire images of each region from the four different conditions, and cFos-immunoreactive nuclei were quantified in Fiji ImageJ (115).

Means ± SEs of the mean were plotted in figures, and statistical testing was conducted in Statistical Package for the Social Sciences (SPSS) (IBM; RRID: SCR_002865↗) and R (RRID: SCR_001905↗). α = 0.05 was adopted in all analyses unless otherwise specified. To analyze behavioral data (locomotor activity phases, immobility-defined sleep, recognition memory performance, and stimulus exploration times), we conducted within-subjects and split-plot ANOVAs with Lighting, Time of Day, and Type of Stimulus as within-subjects factors and Sex and Genotype as between-subjects factors; for locomotor activity under DD, we conducted between-subjects ANOVAs with Prior Lighting as a factor. To analyze molecular data (clock gene expression and cFos+ cell counts), Lighting × Time of Day between-subjects ANOVAs and Lighting × Time of Day × Region split-plot ANOVAs were conducted. Significant interaction effects in most cases were followed up with simple effect analyses in SPSS (116). For main and interaction effects involving within-subjects factors with more than two levels, Greenhouse–Geisser corrections were applied to the degrees of freedom (df) whenever the assumption of sphericity was violated (117); these F values with adjusted df were denoted as F_Ɛ. To examine the effects of sleep history and sample exploratory activity on test performance, we conducted linear mixed-effects models in R (62), with preceding 2-h sleep proportion or sample exploratory activity as the fixed effect, individual mice as the random effect, and recognition scores as the response variable. The significance of the fixed effect was assessed by examining the change in deviance (−2 × maximum log-likelihood) from comparing linear mixed-effects models with versus without the fixed effect of interest. A significant drop in deviance—determined by the likelihood ratio χ² value—indicates that incorporating the fixed effect improves model fitting and explains more variance in the response variable than when it is excluded from the model (118). The 95% CI of the fixed effect (slope) was computed from parametric bootstrap with 10,000 iterations.

Data reported in this manuscript were deposited by S.K.E.T. on Figshare (September 8, 2021) and can be accessed via https://figshare.com/articles/dataset/Tam_etal_data_xlsx/16583834↗. Supplementary methods, tables S1 to S7, and figures S1 to S11 can be found in the SI Appendix↗.