The core components of the molecular clock include the activating transcription factors BMAL1 and CLOCK and the negative regulatory feedback elements encoded by the Per and Cry genes (11, 12). Bmal1 is the only circadian clock gene without a known functional paralog and hence the only one for which a single gene deletion causes a complete loss of behavioral and molecular rhythmicity (13). Because its gene product BMAL1 is a transcription factor that likely influences many downstream genes, Bmal1^−/− mice also exhibit other physiological defects unrelated to the circadian defect (14), including progressive arthropathy, decreased locomotor activity levels and body mass, and a shortened life span (15-18).

In this study, we used Bmal1^−/− mice, which harbor a null allele at the Bmal1 locus (19). The circadian patterns of locomotor activity (LMA) and body temperature (T_b) were monitored by telemetry (Fig. 1, A to C, and fig. S2). As previously reported, these animals showed no circadian rhythms in a 12-hour-light/12-hour-dark (LD) cycle or constant darkness (DD) when given ad libitum (AL) access to food (Fig. 1B). We also attempted to entrain Bmal1^−/− mice to a 4-hour window of RF during the normal sleep period for mice, under both LD [ZT4-8 (from 4 to 8 hours after light onset)] and DD [CT4-8 (from 4 to 8 hours after presumptive light onset)] conditions. In LD and DD conditions, wild-type (WT) and heterozygous littermates showed an elevation of T_b and LMA ~2 to 3 hours before food became available (Fig. 2, A and D, and fig S2). By contrast, Bmal1^−/− mice did not show a comparable elevation in T_b or increase in LMA before the window of RF in DD; T_b and LMA were, however, markedly elevated in the Bmal1^−/− mice after food presentation (Fig. 2, B and D, Fig. 3C, and fig. S2). In addition to the preprandial elevation in T_b under RF, WT and heterozygous littermates demonstrated a clear circadian T_b rhythm (Fig. 2A), whereas Bmal1^−/− mice demonstrated a persisting ultradian T_b pattern throughout the remainder of the day (Fig. 2B). In DD conditions, Bmal1^−/− mice occasionally showed periods of torpor (T_b below 31°C), which were distributed randomly across the circadian day. Consequently, the Bmal1^−/− mice not only failed to show elevation of T_b or LMA in anticipation of the RF but often slept or were in torpor through the window of RF, requiring us to arouse them by gentle handling after presentation of the food to avoid their starvation and death during RF.

After 14 days in this RF regimen, mice were killed to examine clock gene expression in the brain and were compared to mice that had been fed AL. As previously reported (8), WT animals with AL food showed peak expression of Per1 and Per2 mRNA at ZT5-6, and Bmal1 at ZT18-19 in the SCN (Fig. 3, D and E), but little or no expression at other hypothalamic sites. By contrast, WT animals under RF also showed no change in this expression pattern in the SCN (Fig. 3, D and E) but did show induction of Per1 and Per 2 at ~ZT3-9 (preceding, during, and after the RF window) in the DMH with peak expression levels at ZT7-8 (Fig. 3F). We also saw induction of Bmal1 mRNA in the DMH, with peak expression at ZT18-21 (Fig. 3G), consistent with neurons in the DMH showing induction of rhythmic expression of the entire suite of clock genes during RF. As previously reported for Per1 and Per2, Bmal1 gene expression was restricted to the compact region of the DMH. Finally, as expected in Bmal1^−/− mice, no Bmal1 mRNA and very low expression levels of Per1 and Per2 were detected in the SCN and DMH under any feeding or lighting condition (Fig. 1E, Fig. 2, E to G, and fig. S4).

These results suggest that a BMAL1-based circadian clock may be induced in the DMH after starvation and refeeding to drive entrainment of circadian rhythms to the time of food availability. This hypothesis is consistent with previous studies showing that the DMH is a major recipient of direct and relayed input from the SCN and that it is important in relaying circadian signals for sleep-wake cycles, LMA, feeding, and corticosteroid rhythms to other brain systems (20). c-Fos expression in the DMH, but not in the SCN, is shifted to coincide with the activation of LMA and T_b during RF, and lesions of the DMH prevented entrainment of LMA, T_b, and sleep-wake cycles to RF [(3); also see (9, 10)].

To test the role of the DMH-inducible clock in entrainment of circadian rhythms, we attempted to rescue both light and food entrainment of circadian rhythms by injecting adeno-associated viral vectors (AAV, serotype 8) containing the Bmal1 gene (including both 5’ and 3’ promoter elements) (19) into the brains of Bmal1^−/− mice. To test this construct, we first injected AAV-BMAL1 into the SCN. All mice with SCN injections (n = 6) (Fig. 1C), but none in which the injections missed the SCN (n = 16) or in which we injected a different AAV containing the gene for green fluorescent protein (GFP) into the SCN (n = 3), demonstrated entrainment of both LMA and T_b rhythms to a 12:12 LD cycle (Fig. 1C). When the mice were in DD conditions, this rhythm continued as a high-amplitude, free-running rhythm of 23.7 hours (Fig. 1C and fig. S1). Thus, focal bilateral injection of AAV-BMAL1 into the SCN of Bmal1^−/− mice rescued the fundamental properties of the circadian oscillator, including light entrainment, free-running period, and rhythm amplitude. A previous study had shown that rescue of Bmal1^−/− mice could be achieved by a transgene in which BMAL1 was placed under a constitutively active cytomegalovirus promoter (13). However, this gene construct, which was expressed continuously throughout the brain, produced a circadian period of ~22.7 hours, about 1 hour shorter than that of WT mice (13). By contrast, in our study, when we placed BMAL1 under its own promoter and restored this gene only to the SCN, the period of the circadian cycle was precisely the same as in WT animals (Fig. 1, A and C, right panel). Thus the expression of BMAL1 under its own promoter in the SCN alone is sufficient for recovery of light-entrained circadian rhythms. Our results also establish the SCN as sufficient for the generation of the circadian T_b rhythm, a point that has been in dispute (21).

On the other hand, similar to the mice with transgenic replacement of BMAL1 throughout the brain (14), locomotor activity levels in the animals with AAV-BMAL1 injections into the SCN remained significantly lower (P < 0.001) than those of WT and heterozygous littermate mice. Moreover, mice with AAV-BMAL1 injections into the SCN did not show improvement in any of the other physiologic deficits. Hence, these deficits are unlikely to be due to loss of circadian rhythmicity per se (14).

We next tested the animals with BMAL1 replacement in the SCN for their ability to entrain to a restricted temporal window of food availability. Previous studies had demonstrated that animals with lesions of the SCN could still entrain to food, suggesting that there was a food-entrainable oscillator elsewhere in the animals but not excluding participation of the SCN in intact animals (4, 7, 22). When animals with SCN injections of AAV-BMAL1 who had complete rescue of the light-entrained rhythms of LMA and T_b were placed into a food-restriction paradigm under DD conditions, we found that they maintained the rhythm that had been entrained by light (high-amplitude, free-running period of ~23.7 hours) and never showed an increase in LMA or T_b in anticipation of the food presentation (fig. S3). Hence, although a BMAL1-based clock is necessary to support food entrainment, restoration of clock function in the SCN alone is not able to rescue this behavior.

To test the hypothesis that the BMAL1-based clock induced in the DMH during restricted feeding might drive circadian entrainment, we performed stereotaxic bilateral delivery of AAV-BMAL1 (the same construct and vector as used in the SCN) into the DMH of Bmal1^−/− mice. Mice who sustained bilateral DMH injections of the AAV-BMAL1 did not demonstrate entrainment to a 12:12 LD cycle or free-running rhythms of T_b or LMA in DD (Fig. 3A). By contrast, under conditions of food restriction in DD, they exhibited a clear anticipatory increase in T_b and LMA before food presentation (Fig. 2C and Fig. 3B). Each individual mouse showed very little day-to-day variation in the timing of the increase in T_b and LMA under DD (i.e., the phase angle of entrainment was stable). Finally, the increase in T_b and LMA before the predicted period of food presentation persisted during a 24-hour fast at the end of restricted feeding (arrow in Fig. 3B), demonstrating the circadian nature of the response.

In both our study and the study by Mieda et al. (8), clock gene expression in the DMH was largely restricted to cells in the compact part of the nucleus, which consists of small, closely packed neurons that are highly reminiscent of the SCN itself. These neurons appear mainly to have local connections with the adjacent output zones of the DMH (23), suggesting that the timing signal from the compact part of the DMH may impinge upon the same output neurons in the remainder of the DMH as are used to control light-entrained rhythms directed by the SCN. This relationship may explain how the DMH clock is able to override the SCN clock input during conditions of food entrainment in an intact animal. It is unlikely that feedback from the DMH alters activity in the SCN in any major way, because the SCN remains phase-locked to the LD cycle for many weeks during food entrainment (as long as the animals are not also hypocaloric). These observations also raise the interesting possibility that the DMH may form the neuroanatomic basis of the so-called methamphetamine-sensitive circadian oscillator (MASCO), which also operates independent of the SCN and does not entrain to light [for a review, see (24)].

Our data indicate that there is an inducible clock in the DMH that can override the SCN and drive circadian rhythms when the animal is faced with limited food availability. Thus, under restricted feeding conditions, the DMH clock can assume an executive role in the temporal regulation of behavioral state. For a small mammal, finding food on a daily basis is a critical mission. Even a few days of starvation, a common threat in natural environments, may result in death. Hence, it is adaptive for animals to have a secondary “master clock” that can allow the animal to switch its behavioral patterns rapidly after a period of starvation to maximize the opportunity of finding food sources at the same time on following days.

In an intact animal, peripheral oscillators in many tissues in the body, including the stomach and liver, as well as elsewhere in the brain, may contribute to food entrainment of circadian rhythms (25, 26). Consequently, it has been difficult to dissect this system by using lesions of individual components of the pathway (3, 9, 10). However, by starting with a genetically arrhythmic mouse and using focal genetic rescue in the brain, we have identified the SCN molecular clock as sufficient for light but not food entrainment of T_b and LMA rhythms in mice, and the DMH as sufficient for food but not light entrainment of circadian rhythms of T_b and LMA. These results demonstrate the power of viral-based gene replacement in the central nervous system to dissect complex neural functions.

supplementary materials