Time of feeding and the intrinsic circadian clock drive rhythms in hepatic gene expression

Daily Rhythms in Metabolic Regulators Are Primarily Driven by Temporal Feeding Pattern.

Subsequently, we assessed whether the overt rhythm in food consumption accompanies temporal changes in the activity of major metabolic regulators of the liver. We focused on CREB and AKT. CREB is phosphorylated and activated by extended periods of fasting, while AKT is phosphorylated upon insulin receptor activation. Both in turn regulate cellular energy metabolism. We found that levels of the transcriptionally active phospho-Serine 133-CREB (pCREB) (14) and phospho-Serine 473-AKT (pAKT) (15) levels oscillated in all conditions except in cry1^−/−;cry2^−/− mice subjected to ad lib conditions, where feeding rhythms were absent (Fig. 2). While pCREB peak levels coincided with periods of fasting, pAKT levels peaked after the onset of feeding (Fig. 2 A and B). Although the rhythms in pAKT and pCREB in both WT and cry1^−/−;cry2^−/− mice under tRF were comparable, the peak levels in cry1^−/−;cry2^−/− mice were delayed. Thus, these rhythms can be driven by feeding schedule in the absence of a clock, while the circadian oscillator fine-tunes their phases.

Open in a separate window

Fig. 2.

Daily rhythms in AKT and CREB phosphorylation are driven by temporal feeding pattern. Total and phosphorylated protein levels of CREB (A and C) and AKT (B and D) in livers of WT (A and B), and cry1^−/−;cry2^−/− mice (C and D) under ad lib and tRF feeding conditions. Food availability is indicated in red.

Targets of Metabolism-Driven Transcription Factors Show Clock-Independent Rhythmic Transcription.

We used high-density arrays (HDAs) to measure gene expression changes in the livers of WT and cry1^−/−;cry2^−/− mice under ad lib, tRF, fasting, and refeeding conditions (Fig. S2) to test whether metabolic and stress responsive regulators cause rhythmic transcription under physiological feeding conditions. We focused on transcriptional changes of genes known to be regulated by CREB, AKT (FoxO1-mediated), SREBP1/2 (feeding-activated), and ATF6 (a mediator of cellular stress).

The mRNA levels of CREB and FoxO1 targets showed similar expression patterns under different feeding conditions and genotypes. Fifty targets of CREB (16) and 30 targets of FoxO1 (17) were repressed by food intake following prolonged fasting. In WT mice under both ad lib and tRF conditions, transcript levels oscillated, peaking during the respective periods of fasting (Fig. 3 A and B).

Open in a separate window

Fig. 3.

Circadian and ultradian transcription is driven by feeding state controlled transcription factors CREB, FoxO1, SREBP-1/2, and ATF6. Median temporal expression pattern of transcriptional targets of CREB (A), FoxO1 (B), the SREBPs (C), and ATF6 (D) in the liver of WT and cry1^−/−;cry2^−/−.

Feeding after 24 h fasting induced the expression of 31 ATF6 targets and 32 SREBP1/2 targets (18) with distinct kinetics. ATF6 targets were rapidly induced to peak levels 4 h after feeding onset and declined to baseline levels quickly afterward. Such rapid response kinetics of ATF6 targets resulted in weak 12-h transcriptional rhythms under ad lib conditions in WT mice reflecting their bimodal feeding pattern, and they assumed 24-h periodicity under tRF conditions. In contrast to ATF6 targets, SREBP targets displayed a slow induction to about 4.5-fold above fasting levels after 6 h of food availability. Such slow response to feeding and the lack of robust feeding consolidation under ad lib feeding blunts transcriptional rhythms in SREBP targets to almost undetectable levels. Under tRF conditions, however, SREBP targets showed rhythmic expression patterns with peak phases attained at 6–8 h after feeding onset (Fig. 3 C and D).

In cry1^−/−;cry2^−/− mice under ad lib conditions, the targets of the above described transcription factors were arrhythmic. Under tRF conditions, these transcripts showed clear 24-h expression rhythms similar to expression in WT mice (Fig. 3). However, the peak phases of transcript expression occur slightly earlier in WT mice than in their cry1^−/−;cry2^−/− mice counterparts, suggesting that the oscillator may potentiate transcriptional responses of these metabolic regulators to feeding and fasting.

The observation that four different classes of metabolic/stress responsive transcription factors drive liver-clock-independent rhythmic transcription raised the question, whether all food-responsive transcripts show clock-independent rhythmic expression reflecting feeding behavior.

Feeding-Responsive Transcripts Show Ultradian and Circadian Transcription Following the Pattern of Food Intake.

To identify transcripts that are acutely affected by feeding and to minimize any confounding effect of light and the circadian clock on the fasting/feeding modulated transcription, we subjected WT mice held under DD to a full circadian cycle (24 h) of fasting starting from CT4 before refeeding. This length of fasting has moderate effect on hepatic gene expression (19) and does not arrest the intrinsic clock (3). Since transcriptional changes in response to refeeding can have different kinetics (20), we monitored expression every 2 h for three time points after feeding. As previously defined (15), we set a 1.6-fold change threshold criteria and identified 1,089 food-induced and 992 food-repressed probe sets. Heat map rendering of the 24-h expression profile of these probe sets in WT or cry1^−/−;cry2^−/− mice showed a clear rhythmic expression pattern reflecting feeding behavior (Fig. 4). The rhythmic transcription of feeding-modulated genes is well pronounced in the median of their normalized expression profiles (Fig. 4 B and D). In WT mice under ad lib feeding conditions, many food-induced probe sets showed synchronous, low-amplitude rhythms with elevated levels during the nighttime and two peaks, coinciding with the two nocturnal feeding bouts (Fig. 4 A and B). Under tRF conditions, as feeding was consolidated to one bout, feeding regulated transcripts displayed distinct 24-h rhythms in WT and cry1^−/−;cry2^−/− mice, with peak levels of food-induced transcripts following food intake and food-repressed transcripts preceding food intake (Fig. 4).

Open in a separate window

Fig. 4.

Rhythmic food intake drives rhythmic gene expression. Heatmap rendering (A and C) and median expression changes (B and D) of food-induced and food-repressed transcripts in the indicated genotypes and feeding conditions. (A and B) Columns indicate time points, and rows indicate single transcripts. High levels are shown in yellow, whereas low levels are shown in blue. Data for WT are at 1-h resolution, while data for cry1^−/−;cry2^−/− mice are at 2-h resolution. Median expression values were smoothed by a 3-h moving average.

Global Transcriptional Rhythms in tRF fed cry1^−/−;cry2^−/− Mice Can Be Partly Restored by Food Intake.

The heatmap rendering of feeding-modulated probe sets showed that feeding consolidation can drive rhythmic gene expression in cry1^−/−;cry2^−/− mice into two principal phases coinciding with time of feeding and fasting (Fig. 4). This prompted us to assess the extent of global rhythms in gene expression driven by an imposed feeding rhythm in the oscillator deficient cry1^−/−;cry2^−/− mice. Hepatic transcription profiles of cry1^−/−;cry2^−/− mice under ad lib or tRF feeding conditions were analyzed to find transcripts with robust approximately 24-h periodicity (21). As shown earlier (10), no significant rhythm in gene expression was found in ad lib fed cry1^−/−;cry2^−/− mice (Table S1). However, feeding consolidation restored rhythmicity for 617 probe sets. These probe sets were clearly split into two groups, one consisting of food-induced transcripts and the other consisting of food-repressed transcripts (Fig. 5). The probe sets also showed 24-h rhythms in WT mice under tRF conditions with comparable, but not identical phases of expression (Fig. 5). Additionally, we found that nearly all transcripts previously shown to be rhythmic in the liver of liver-specific Bmal1 knockdown mice (12) gained rhythmic expression in cry1^−/−;cry2^−/− mice liver under tRF (Fig. S3).

Open in a separate window

Fig. 5.

Temporally restricted feeding can partially restore approximately 24-h rhythm in gene expression in the livers of cry1^−/−;cry2^−/− mice. Heatmap rendering of temporal gene expression pattern of 617 probe sets scored rhythmic in cry1^−/−;cry2^−/− mouse liver under tRF.

Plasticity of Circadian Transcriptome Under Different Feeding Conditions.

The effect of feeding and fasting on temporal gene expression suggested that the repertoire, phase, and amplitude of rhythmic transcripts in the WT liver are products of both feeding pattern and the intrinsic circadian clock.

Feeding consolidation under tRF condition improved rhythms in a set of 3,217 probe sets, while dampening the rhythms of a set of 1,254 (Fig. 6A and Fig. S4 A and B). Many oscillator components continued rhythmic expression under both conditions.

Open in a separate window

Fig. 6.

Feeding schedule affects rhythmic transcription. (A) Transcripts that were scored as rhythmic in WT mice under both ad lib and tRF or fasting condition are shown in a Venn-Diagram. (B) Heatmap rendering of temporal gene expression pattern of the 1,743 “common cycler” transcripts is shown. (C) Venn-Diagram depicting only 368 of 2,997 transcripts rhythmic in ad lib-fed WT mice maintained a robust transcriptional rhythm under fasting conditions. (D) Heatmap rendering of these 368 transcripts, including most core clock components and known clock-controlled genes under ad lib and fasting conditions.

Also, nearly 28% of transcripts rhythmic in both conditions (common cyclers) were induced or repressed by a factor of 1.6-fold upon refeeding. Heatmap rendering of their expression patterns (Fig. 6B) revealed effect of feeding on the expression of many of the remaining common cyclers. Accordingly, feeding pattern also influenced the phase and amplitude of oscillation of most of the common cyclers. When feeding was changed from the predominantly nocturnal ad lib to daytime tRF conditions, the vast majority (84%) of common cyclers showed >6-h shift in their phase of expression (Fig. 6A and Fig. S5). Furthermore, >73% of the common cyclers, including the core oscillator components and their immediate targets, showed an increase in amplitude of oscillation (Fig. S5 A–D).

Most of the “common cyclers” group of transcripts were not scored rhythmic by the applied filters in cry1^−/−;cry2^−/− mice under tRF conditions. However, many showed a low amplitude rhythmic transcription pattern with phases similar to those in the WT mice (Fig. S3B). Therefore, feeding consolidation may generate low amplitude rhythms, however a functional hepatic clock may fine tune their phases (Fig. S6) and make them more robust (example in Fig. S7). Interestingly, we identified only a small group of nine transcripts displaying similar expression patterns, which did not change their phase of peak-expression between the two feeding conditions (Fig. S5 F and G). We hypothesized that these transcripts might be driven by systemic cues originating in the SCN, which is unperturbed by feeding pattern (3, 4).

Food Driven Transcripts Fail to Sustain Their Rhythms in the Absence of Food Intake.

The circadian oscillations of the core clock mechanism are sustained during fasting of up to 48 h (3, 22). Output transcripts whose expression is dominantly driven by the hepatic circadian clock should also sustain their oscillation under fasting conditions.

In separate experiments, we subjected WT mice to approximately 24-h fasting periods starting from CT4 or CT14 and performed gene expression profiling at 2-h intervals for up to 10 h starting at CT2 or CT12, respectively (see Fig. S2 for experimental outline). Covering only 22 h, the sampling protocol may slightly under-report the number of rhythmic transcripts.

The analyses revealed that >80% of transcripts rhythmic under ad lib condition cease to oscillate when food is withheld, while only 368 of the 2,997 rhythmic transcripts including the core oscillator components continue self-sustained rhythms in the absence of feeding (Fig. 6 C and D). This suggested that the perturbed metabolic state under prolonged fasting does not stop the circadian oscillator, but blunts rhythmic expression of majority of hepatic transcripts. Hence, food intake is a major cause of rhythmic transcription even in the presence of a functional clock.

Food Modulated Transcripts Constitute a Metabolic Sand Timer.

A large number of hepatic transcripts show increased expression upon feeding or fasting. Food-induced transcripts rise following daily physiological feeding bouts, while the fasting-induced transcripts accumulate with increasing length of fasting. Many of these feeding/fasting-regulated transcripts are direct targets of metabolic and stress regulators CREB, SREBP, ATF6, and FoxO1. The cyclic activities of these metabolic regulators loosely coupled to the circadian clock, but a daily rhythm in feeding can drive their rhythmic activities even in the absence of a clock. Conversely, in the absence of feeding, a functional clock cannot impose rhythmic expression of their targets. Hence, the food-driven transcripts are analogous to a metabolic sand timer as opposed to the self-sustaining rhythms of a circadian oscillator (Fig. S8).

Prolonged Fasting Suppresses Rhythmic Transcription of a Large Number of Transcripts.

Under fasting conditions, only 368 probe sets continued to oscillate robustly (Fig. 6 C and D and Fig. S4C). This small subset includes the known core oscillator components and their well-characterized immediate targets. In the absence of food intake, most food-driven transcripts fail to sustain circadian variations, highlighting their nature as a response to and not in anticipation of feeding. The cellular levels of various products of energy metabolism, such as NADH and ATP, which are directly used in transcription or are necessary cofactors for various chromatin modifying enzymes, drop during low nutrient states (27–29). Therefore, prolonged fasting may globally suppress dynamic changes in transcription. However, resistance of the circadian oscillator to severe perturbations of general transcription (30) may allow sustenance of the core clock under prolonged fasting.

Down-regulation of circadian transcriptional outputs under low nutrient state may be a universal phenomenon across kingdoms. In plants, up to 30% of the transcriptome, including several enzymes and regulators for nutrient production and metabolism, shows rhythmic expression under constant light when photosynthesis persists (31, 32). However, under DD only a handful of transcripts including the components of the oscillator and their immediate outputs show sustained rhythms (33). This may be a mechanism to conserve energy by preventing wasteful synthesis and degradation of a large number of transcripts.

Circadian Oscillator Modulates Expression of Food-Driven Transcripts.

In the absence of a circadian oscillator in cry1^−/−;cry2^−/− mice, scheduled feeding can rescue rhythmic expression of only 617 food-driven transcripts. A large number of food-driven transcripts do not attain the same level of amplitude as they do under similar feeding conditions in the WT mice, thus are not detected as rhythmic in cry1^−/−;cry2^−/− mice. This highlights the role of an oscillator in enhancing the amplitude of oscillation in food-driven transcripts.

Under identical tRF conditions, the peak phases of expression of feeding driven transcripts in WT mice occur 1–2 h earlier than in cry1^−/−;cry2^−/− mice (Fig. S6). This delay of peak expression in cry1^−/−;cry2^−/− mice does not arise from changes in the feeding pattern as reflected in their RER (Fig. 1). A possible explanation is that the circadian oscillator anticipates changes in the feeding state and accelerates the transcriptional response to an acute activation or repression by feeding cues in food-dependent genes. Similar oscillator-mediated potentiation of rhythmic posttranslational modifications is also observed for CREB and AKT phosphorylation. In summary, although an anticipatory role of the circadian clock has long been hypothesized, here it has been demonstrated in genome-wide transcriptional regulation.

Direct Effect of Feeding on the Circadian Clock.

Among various circadian clock components, the mRNA level of Per1 is down-regulated, while Per2 is up-regulated by food input, causing their peak expression to occur before or after the feeding onset, respectively (Fig. S8A). It is well known that feeding-induced changes in CREB and heat shock factor (HSF) activities can regulate Per1 and Per2 expression, respectively, from their target binding sites present in Per1 and Per2 promoters (34, 35). The acute dephosphorylation and consequently inactivation of CREB (Fig. 2A) following feeding may down-regulate Per1. Postprandial rise in body temperature, on the other hand, may activate Per2 transcription via HSF binding to the cis-acting element in the Per2 promoter (12). These mechanisms are independent of a functional clock, as the differential effects of refeeding on Per1 and Per2 mRNA levels are preserved in the cry1^−/−;cry2^−/− mice.

Interaction between Clock and Temporal Feeding Pattern Determines the Circadian Transcriptome.

Rhythmic feeding behavior increases the amplitude in the oscillations in core clock components and their immediate target genes, thereby causing robust rhythmic transcription (Fig. S5 A–D). In turn, the circadian clock modifies and enhances the transcriptional response to feeding cues. As a result, 4,960 transcripts display rhythmic expression in WT mice under consolidated feeding (tRF). This largely surpasses the combined numbers of food-only (617 transcripts are rhythmic in cry1^−/−;cry2^−/− mice under tRF conditions) and circadian clock-only (368 rhythmic transcripts under fasting condition in WT mice) driven transcripts, even when lower sampling frequency is taken into account (26). This synergistic interaction between clock and metabolism is exemplified in the subset of rhythmic transcripts that encode the mitochondrial components (Fig. S7). In WT mice under tRF conditions, these transcripts exhibit more coherent phases and robust rhythms in expression than WT mice under ad lib conditions or cry1^−/−;cry2^−/− mice under tRF conditions.

Animals.

All animal experiments were carried out in accordance with the guidelines of the Institutional Animal Care and Use Committee of the Salk Institute. Male wild-type C57BL/6J mice (8- to 12-week-old) and 8- to 16-week-old cry1^−/−;cry2^−/− mice were used. Metabolic data were measured by indirect calorimetry.

Real-Time Quantitative PCR.

Total RNA was pooled from at least three livers and used for cDNA synthesis. Transcript levels were then quantified by qPCR using the SYBR Green PCR Master Mix (Applied Biosystems). Results for the respective gene of interest were normalized to Gapdh and subsequently median-normalized to 1.

Western Blot Analysis.

Proteins were detected using antibodies against pCREB (37), CREB (38), pAKT (#7100; Cell Signaling), AKT (#9272; Cell Signaling), and βACTIN (A2066; Sigma–Aldrich), which were used at the dilution proposed by the manufacturer.

Probe Hybridization and Data Analysis.

Total RNA was pooled from three liver samples and amplified, labeled, and hybridized on Affymetrix MOE 430_2 HDAs using the One-Cycle Amplification kit (Affymetrix) following the manufacturer's protocol. Raw data were analyzed in R. Rhythmic transcripts were identified using COSOPT and Fisher G test as described in refs. 11 and 39. Heatmaps were generated with the Cluster and Treeview (40) program.

The authors thank Aziz Sancar for the cry1^−/−;cry2^−/− mice. This work was supported by the National Institutes of Health (NIH) Grant NRSA 1F32GM082083–01 (to L.D.) and the Pew Scholars Program in Biomedical Sciences and by NIH Grant EY016807 (to S.P.).