Eating around the clock: circadian rhythms of eating and metabolism

Abstract

The time of day that we eat is increasingly recognized as contributing as importantly to overall health as the amount or quality of the food we eat. The endogenous circadian clock has evolved to promote intake at optimal times when an organism is intended to be awake and active; however, humans have created an environment in which whenever their eyes are open, their mouths are also typically open and eating. In this review, we highlight literature pertaining to the effects of food timing on health, beginning with animal models and then translation into human experiments. We emphasize the pitfalls and opportunities that technological advances bring in bettering understanding of eating behaviors and their association with health and disease. There is great promise for restricting the timing of food intake both in clinical interventions and in public health campaigns for improving health via non-pharmacological therapies.

1. Introduction

Circadian rhythms evolved so that organisms could anticipate predictable changes in the environment brought about by the path of the Sun in the sky. These rhythms are endogenous, meaning that they continue to cycle without external cyclical input. Circadian rhythms are approximately 24 h long (circa – dia, “about a day”) – these act in cells to ensure that physiology and behavior can adapt and synchronize to external cycles of light, temperature, and food availability, among other factors. In mammals, the circadian control system is distributed across tissues, with a central pacemaker in an area of the hypothalamus called the suprachiasmatic nucleus (SCN). This area receives direct projections from the retina and it in turn synchronizes clocks throughout the rest of the brain and body through both neural and humoral factors (41). These extra-SCN clocks are also sensitive to nonphotic cues such as those related to eating (Figure 1) (41, 94).

Figure 1.

Central and peripheral clocks are entrained by light and food-related cues. Central and peripheral clocks interact with each other and with cues associated with behavior and metabolism. Darker arrows indicate stronger effects. Food timing has variable effects on central clocks. Food timing is a weak entraining cue for the SCN, but is stronger for other extra-SCN brain clocks (94).

When peripheral clocks are mis-timed with the central clock due to feeding at an inappropriate time, the mismatch in timing can perturb the normal coupling of circadian clocks and result in grave consequences for health. Epidemiological studies indicate that shift work and social jetlag (both defined in Terms and Definitions) are associated with a number of adverse outcomes including obesity, diabetes, and cardiovascular disease (53, 108, 128). One common thread in circadian disruption is the impact of altered food schedules. Greater food intake during the biological rest hours (night for diurnal animals and vice versa) is associated with many of the same adverse cardiometabolic outcomes as occur with other circadian disruptions (17, 18, 30). Observational and interventional studies in human participants, as well as laboratory studies of model animals, underscore the adverse effect of mis-timed food (10, 123).

Circadian disruption occurs in numerous ways. An organism’s circadian phase (timing of the rhythm) can be misaligned, either because the behavior is misaligned with phase (e.g., shift work or jetlag), or because phase is misaligned with the external schedules (e.g., chronic misalignment caused by advanced or delayed sleep phase syndrome). Circadian phase can also be shifted to new and inappropriate times by external cues (e.g., light at night delaying the clock). In addition to phase disruptions, an organism’s natural clock speed (free running period, τ) can differ enough from the environmental cycle such that the clock free-runs with respect to the environment. This is unusual but occurs in people that are completely blind and thus unable to entrain daily to the light/dark cycle, as the average circadian period in humans is longer than 24h. These different circadian disruptions are not mutually exclusive, and they can combine in synergistic ways. One example is social jetlag, which is the experience of early awakenings on work-days and late awakenings on free-days. This means that the body clock is often misaligned with the environment, and moreover may be dynamically trying to resynchronize each week.

Clock disruption has numerous adverse outcomes, but what is the value of a working clock and does it promote health when stably entrained and aligned? Pittendrigh and Bruce (111) first posited the circadian resonance hypothesis. Resonance is defined by a match between the endogenous period and the driving T-cycle (τ = T). A match between τ and T improves health and fitness across a wide range of taxa. For example, fitness is improved in cyanobacteria (105) and longevity is enhanced in animals ranging from fruit flies to rodents to primates (55, 84, 112). Similar beneficial results of resonance are observed with respect to metabolism (discussed below). These examples consider resonance of circadian period with the light-dark cycle only. But organisms live in complex environments in which multiple biotic and abiotic cues may cycle, and humans in particular live in temporal environments in which both imposed schedules and self-selected behaviors may cycle in time with each other or not. In the last two decades, substantial advances have been made by exploiting food as a synchronizer of clocks, in the form of time-restricted food access to improve health (generally termed time-restricted feeding (TRF) or time-restricted eating (TRE) for rodents and humans, respectively) (81). The value of TRF/TRE may rest in its support of circadian resonance through food-related cues.

If circadian resonance is health-promoting, what is the particular physiological stress that is associated with circadian disruption? And are all forms of circadian disruption acting on similar pathways? For example, both internal and external misalignment involve periodic stimuli or stressors acting on clocks at the “wrong” time. This may engage similar or different effector pathways in different tissues. The chronic nature of circadian disruption also occurs continuously (advanced/delayed phase), regularly (social jetlag or shift work), or intermittently (jetlag for frequent travelers). For a discussion of circadian disruption mechanisms, see Vetter (146).

For this review, we explore the particularly pernicious effects of mistimed eating on health and how resonance of food timing, light-cycle timing, behavior, and internal clocks impact health and disease.

2. Circadian physiology. A distributed multi-oscillator system

Circadian clocks have evolved multiple times, likely as an adaptation to partitioning energy utilization, growth, and repair to specific times of day when resources (sunlight, temperature, food availability) make this most efficient (60). Clocks allow organisms to be most active when likely to find food and encounter mates and to be least active during the costly metabolic times (usually at night) or when predation pressure is strongest (more often in the day). In vertebrates, the circadian clock is a multi-oscillator system (Figure 2). A central pacemaker lies in the SCN of the hypothalamus, situated above the optic chiasm from which it receives light information from the retina. The SCN was long considered the critical time keeper because ablations of this area rendered animals arrhythmic in their activity, food intake, and drinking (98, 136). Yet food was still a key towards recognizing that circadian rhythms did not begin and end with the SCN but instead permeated physiology. When presented with daily food access, otherwise arrhythmic SCN-lesioned animals could still predictably and reliably anticipate the food time (food anticipatory activity) (135). The later discovery of clock genes that functioned throughout the body would provide a mechanism to help explain the lasting 24h rhythmicity in animals that we thought lacked a timekeeping system.

Figure 2.

Circadian and appetite regulation in the brain. Main areas associated with circadian timing are shown with their projections to hypothalamic and brain stem areas important for regulating eating. Right inset shows the core transcription-translation feedback loops of the molecular circadian clock. Suprachiasmatic nucleus (SCN), subparaventricular zone (SPVZ), paraventricular nucleus (PVN), lateral hypothalamus (LH), dorsomedial hypothalamus (DHM), arcuate nucleus (Arc), parabrachial nucleus (PB). Inset adapted with permission from the American Thoracic Society. Copyright © 2016 American Thoracic Society. All rights reserved. Balachandran et al. (14). Annals of the American Thoracic Society is an official journal of the American Thoracic Society.

2.1. Clock genes, central and peripheral

The SCN is composed of autonomous single cell oscillators that are coupled with one another to create a network that can synchronize to zeitgebers (“time givers”). Network function depends on the strength of coupling among individual oscillators and their underlying connectivity structure (153). The resulting ordered rhythms across time and space in the SCN help determine important clock functions, including day length encoding, entrainment, and coordinating outputs to peripheral clocks (44, 51, 58). Within cells, rhythms are based on the action of a transcription-translation feedback loop (Figure 2 inset). This has been well reviewed elsewhere (34). Briefly, in mammals, the loop is comprised of positive elements CLOCK and BMAL1 that heterodimerize and stimulate the transcription of negative elements PERIOD (PER) and CRYPTOCHROME (CRY). The PER and CRY clock gene products heterodimerize themselves, enter the nucleus, and inhibit CLOCK/BMAL1 transactivation. A second loop with REV-ERBα as the key player maintains BMAL1’s rhythmicity. A host of other factors ensure the proper chaperoning of these proteins, and transcriptional control is fine-tuned by other rhythmic epigenetic mechanisms (104).

Not surprisingly, mutations of the core clock genes alter the rhythms of organisms. To date, Bmal1 is the only single gene whose knockout renders animals completely arrhythmic (24). The loss of other genes has more subtle effects, likely because of compensation by paralogs. For Per and Cry, each with two paralogs, single gene manipulations alter the free-running period, but arrhythmicity only follows loss of both Per1 and Per2 or the loss of both Cry1 and Cry2 (13, 145).

Molecular circadian rhythms have now been characterized in virtually all tissues of the body (150, 161). This suggests that the primary role of entrainment is not just to synchronize a central clock to the environment, but rather to maintain local control of clock phase across the body. When considered broadly, approximately 15% of all genes are transcribed rhythmically, but the identity of the cycling transcripts varies across tissues, presumably to support tissue-specific functions (161).

Thus, the hierarchical organization of the body’s clock stretches from organs to cells to sets of rhythmically expressed genes, and each step of this hierarchy allows room for synchronizing inputs. This may explain the difference in sensitivity of oscillators to different cues (e.g., light versus food), and provide a mechanistic basis on which to understand how the disruption of internal rhythms and the misalignment of those rhythms with the environment can cause pathology.

2.2. Circadian control of feeding behavior

Most aspects of physiology and behavior are under circadian control, and most of these strongly covary in animal models, with locomotion, eating, and drinking occurring during wakefulness. These different behaviors are notoriously difficult to separate in rodent models, which is one reason for the power of circadian studies in human participants where these different behaviors can be isolated (Section 4.1). SCN outputs to key behaviors have been reviewed elsewhere (133), and identifying the relevant neurohumoral pathways and molecular mechanisms is an active area of research in chronobiology. Here we briefly consider the circadian control of feeding.

Appetite exhibits a circadian rhythm leading to greater food intake during the active phase. Rodents often exhibit two small peaks of intake around lights-on and lights-off, whereas in humans, appetite is lowest in the morning and increases through the day (116, 124). The consequent eating has profound effects on the circulating metabolic milieu and on peripheral clocks. The hypothalamus integrates acute ingestion-related cues and chronic energetic information such as adiposity-related cues. These brain areas respond to circulating glucose, insulin, leptin, and ghrelin, and the circuits are further modulated by internal processes including stress, arousal, and the circadian clock (137). Important neuron populations and brain areas involved in signaling appetite, feeding behavior, and satiety include: 1) the orexigenic AgRP neurons of the arcuate nucleus where they are opposed by the satiety cues coded by POMC neurons, 2) orexin neurons in the lateral hypothalamus, an area associated with feeding, reward, as well as arousal and wakefulness, 3) the dorsomedial hypothalamus (DMH) that is a major integrating area for food intake and energy homeostasis, and two nuclei that signal satiety, 4) the paraventricular nucleus, and 5) the parabrachial nucleus. The circadian system is intimately linked to these areas to modulate hunger and ingestion over the course of the day (31).

There is a strong efferent projection from the SCN to the paraventricular nucleus, DMH, and arcuate nucleus with sparser projections to the lateral hypothalamus (2, 74) (Figure 2). Additionally, via a major downstream relay of the SCN, the subparaventricular zone, timing information is also conveyed through projections to the lateral hypothalamus, arcuate, and parabrachial nucleus (147). In organizing rhythmicity of feeding, the DMH, paraventricular nucleus, and lateral hypothalamus are of interest because time-restricted access to food can synchronize neuronal activity in these areas (9). Additionally, ibotenic lesions of the DMH can attenuate the night:day ratio of food intake based on weighing of food every 12 h (32). Though this points to the DMH as a key node in organizing food intake rhythms, this result is not always obtained and there may be an important reciprocal connection between the SCN and the DMH in the control of feeding (3, 78). We further note that much of the literature on rhythms and food has focused on food anticipatory activity (locomotor activity) and not specifically food seeking or consumption.

The appetite/satiety coding brain regions also express circadian clocks (1, 31). These clocks play a functional role, because the loss of Bmal1 in the forebrain alone alters food intake patterns (30). For example, AgRP neurons do not simply indicate energetic state homeostatically. Clock genes are autonomously rhythmic in arcuate AgRP neurons and these participate in modulating the cell’s response to leptin (30), and the activity of AgRP neurons is strongly rhythmic and depends on input from the SCN (122). This provides a mechanism for the SCN’s control of daily rhythms in appetite. Nevertheless, nutrient signaling can also affect the clocks in these cells, perhaps explaining how restricted feeding can uncouple rhythms in the brain from rhythms in the SCN. In a single cell sequencing study of AgRP neurons in response to food deprivation, clock genes and clock-related gene ontologies were highly represented in the differentially expressed gene set (52). Therefore, projections from the SCN may be important in two ways. First in altering the balance of excitatory/inhibitory tone in regions associated with energy balance, appetite, and satiety. And second in synchronizing local circadian clocks in these regions.

3. Circadian control of metabolism and eating behavior in animals

3.1. Lessons from clock mutant mice and their eating time

A number of mice with mutations in the core clock genes exhibit metabolic abnormalities, though not all are in the direction of worse metabolic health. For an excellent review and table of phenotypes, see Zarrinpar et al. (160). Mutations of all the main clock genes are associated with change in glycemic or other metabolic control, but it is difficult to parse the relative contributions of alterations in the cellular clock, changes in circadian alignment within the body, timing-independent pleiotropic effects of the genes in question, and behavioral changes like activity, sleep, and eating that would impact energetics. There have been some investigations of food intake in clock gene mutant mice, but not many given the difficulties of measuring food (7) (See section 3.5).

The first clock gene mutation reported to associate with metabolic syndrome was ClockΔ19. Because of this dominant negative allele, the mice have a longer circadian period and they spontaneously develop obesity and exhibit hyperglycemia and hypoinsulinemia (144). The ClockΔ19 mutation impairs insulin secretion (82), but it also blunts their food intake rhythm (129, 144); to what extent these two changes contribute to the final phenotype is an open question.

There are several other examples of clock gene mutants in which a disrupted food intake rhythm is associated with obesity. Mutations of PER1 can exaggerate high fat diet-induced obesity, but this only occurs if the eating rhythm shifts. Liu et al. (80) generated mice with human PER transgenes (hPER) that incorporated the single serine to glycine substitution that has been associated with human familial advanced sleep phase syndrome (hPER2^S662G and the paralogous hPER1^S714G). In the hPER1^S714G mutant, the food intake rhythm is advanced so that intake is evenly split between the light and dark phase and these mice gain much more weight on high fat diet than wildtype mice. In contrast, the hPER2^S662G mutation did not alter food intake and the mice did not gain any more weight on high fat diet than their wildtype controls. A flat day-night ingestion pattern occurs in mPer2 knockouts. Though not sufficient to alter body weight on normal chow, the knockout appears to increase their metabolic susceptibility, for on a high fat diet, they eat more during the light phase than the dark and become obese compared to controls (159). Finally, double knockouts of both Per1 and Per2 do not gain excess body weight. In this study, it was reported that the food intake pattern was arrhythmic, but this was studied in constant darkness after the experiment had been conducted in a light-dark cycle (5). The intake pattern in light-dark is not known, but it might be rhythmic and restricted to the dark if intake follows activity. These double knockouts have a daily rest-activity cycle that is driven directly by light and that is immediately lost in constant darkness (162).

Data on food intake in Cry knockouts is sparse. In Cry1/Cry2 double knockouts, feeding rhythms have been reported to be either arrhythmic (16) or normal (57) though the data were not shown in the latter. Barclay et al. (16) observed that both regular and high-fat chow were eaten arrhythmically, but the knockouts only gained more weight on the high-fat diet. As for the Per2 knockout above, a change in the food intake rhythm alone is not sufficient to induce metabolic dysfunction, but it appears to make the mice more susceptible to an additional metabolic stressor.

Finally, Bmal1 deserves consideration as the only single gene whose knockout causes arrhythmicity. The global Bmal1 knockout mouse exhibits a range of maladies including arthropathy, early aging, a poor ability to utilize fat as a substrate, and though they initially gain more weight than wildtype mice when young, their body weight eventually drops (23, 65, 71, 77). Its feeding rhythm is also arrhythmic (76) but how much this contributes to the phenotype is not known. Bmal1 appears to be a critical piece in metabolic control, for metabolic phenotypes, in particular changes in insulin signaling, emerge in otherwise healthy-looking mice when Bmal1 is knocked out tissue-specifically (77, 82, 125).

3.2. Lessons from circadian disruption by light cycles

Non-24h light cycles or jetlag-style shifting light schedules can have effects similar to those caused by genetic disruptions of the circadian clock (Figure 3). In wildtype mice, a non-resonant T-cycle can increase body mass, elevate glucose, and impair insulin sensitivity (67). The reverse is also true: restoring resonance can be protective. Adlanmerini et al. (6), in studying SCN-specific knockouts of Rev-erbα and β, found excessive weight gain on a high fat diet. This, however, appeared to be due to a change in circadian period and a loss of resonance. They have a short free running period (τ ~ 21 h) and only gained excessive weight when housed in a standard 24h T-cycle. When the mice were housed on a T-21 cycle that matched their τ, the excess body weight gain was prevented and other metabolic measures were ameliorated (Figure 3a). This strongly indicates that the circadian disruption was indeed driving the metabolic disruption as opposed to other pleiotropic effects of the loss of Rev-erb. Even small deviations in resonance may have significance. Wildtype mice gain excess body fat in diet induced obesity on a standard 24h light-dark cycle, but this is prevented when they are housed on a 23.7h cycle that matches their endogenous period (134). To what extent resonance or lack thereof contributes to metabolic abnormalities in the clock mutant mice discussed above remains unknown.

Figure 3.

Different methods to assess the effects of circadian rhythms on body weight. (a) T-cycles of light that are in resonance with the free-running period protect against diet-induced obesity. The actograms show the activity patterns of mice in a short light-dark cycle (10.5h each of light and dark). Each line across shows 2 cycles of activity with vertical hatch marks showing the time and intensity of activity. Yellow indicates lights-on. The SCN-specific Rev-erb knockout (τ~21h) entrains to the 21h T cycle, but the wildtype free-runs without entraining. On a high fat diet, only the mice in a non-resonant T cycle gained excess weight (*significantly heavier than mice in the two resonance groups, adapted from ref. 6). Adapted from Adlanmerini et al. (6) © The Authors, some rights reserved; exclusive licensee AAAS. Distributed under a CC BY-NC 4.0 license. Reprinted with permission from AAAS. Actograms kindly provided by Drs. Marine Adlanmerini and Mitch Lazar. (b) Jetlag. Grey shading in actograms shows the dark phase. In the jetlagged group, they experienced a 6h advance of the light cycle twice per week (right-hand plot). **Weight gain was significantly greater in the jetlagged group (p<.001, from ref. 101). Reprinted from Biochemical and Biophysical Research Communications, Vol 465, Oike et al. (101), Time-fixed feeding prevents obesity induced by chronic advances of light/dark cycles in mouse models of jet-lag/shift work, pp. 556–561, © 2015, with permission from Elsevier. (c) Time-restricted feeding. Left plot shows the time of food access in the light dark cycle; subscripts indicate the duration of food access. Mice that ate high fat diet in the light phase gained significantly more body weight than those that ate in the dark (middle, from ref. 10, with permission, © 2009, The Obesity Society) * Significant difference in body weight. On regular chow, rats also gained significantly more weight when fed during the light phase (data plotted from ref. 121). ** TRF-light rats gained significantly more weight over the experiment.

Chronic jetlag models, such as large phase shifts imposed either weekly or biweekly, can also be a metabolic stress (Figure 3b). Unfortunately, eating behavior has not been well studied in these types of experiments. We return to this in section 3.4 as timed eating can stave off the deleterious effects of jetlag.

Finally, even under static light-dark cycles, small changes can have big effects. Dim light at night is an issue of public health relevance and it can change how behaviors are synchronized to the day in both humans and animals (29, 149). In mice, dim light at night redistributes food intake into the normal rest phase, and the final body weight gain is correlated with the calories consumed at the inappropriate time (45). Importantly for translation, dim light at night impairs glucose tolerance in both nocturnal and diurnal rodents (85, 103) and humans (86).

3.3. Lessons from circadian disruption by food timing and diet composition

When animals eat can have as much effect on metabolic health as what they eat: mistimed food reliably compromises metabolism (10, 21, 45, 121, 127) (Figure 3c). The power of food timing to organize behavior has been evident for a long time. It is also a potent zeitgeber that can synchronize circadian rhythms in extra-SCN brain areas and other peripheral organs. In nocturnal rodents that eat during the light phase, the light cycle entrains the phase of the SCN, but the food cycle entrains rhythms in most other areas of the body (37, 138). Food timing also entrains neuronal activity rhythms in extra-SCN brain areas (9). Food, when presented periodically, is such a powerful organizing cue, that this has led to the broad adoption of a model in which the SCN controls feeding, and the ingestion-related cues thereafter entrain the peripheral clocks (126).

In TRF, the fasting interval appears to be critical to entrain peripheral clocks. Fasting in mice quickly causes significant hypoinsulinemia. The drop in insulin concentration leads to phosphorylation of glycogen synthase kinase 3β and increased PPARα, and the two of these ultimately elicit untimely increases in the transcription of clock genes Rev-erbα, Per1, and Per2 (97). In addition to resetting by low insulin, the sharp increase in insulin with refeeding can also reset the clock (158). Insulin can induce PER expression (35) but can also phosphorylate BMAL1 via Akt to reduce its accumulation in the nucleus (38).

With food, what and when can interact, so nutrient content can influence the timing of eating. For example, when mice are switched from a standard to a high fat diet, the intake pattern shifts to one of more consistent grazing over 24 h (70). This change in intake pattern is mirrored by the activity pattern, and it is accompanied by a phase shift of the liver but not of other tissues (109). A similar reduction in circadian amplitude of activity occurs in rats on high fat diet (20). Surprisingly, this effect of high fat diet on eating behavior is sex-dependent and the blunted rhythm of intake occurs only in males or in females that are ovariectomized. This indicates an important modulating role of estrogens on circuits involved in eating and that high estradiol permits a high amplitude circadian rhythm to be preserved (106). Finally, the functional importance of the eating pattern is underscored by the observation that mouse strains that are resistant to diet induced obesity are better able to maintain normal high amplitude feeding rhythms on high fat diet (22).

We have focused on the impacts of high fat diet, as this is the most-thoroughly studied and prevalent nutritional manipulation in studies of metabolism and the clock. Though beyond the scope of this review, macro-and micro-nutrients including fat, cholesterol, sugars, amino acids, vitamins, polyphenols, and phytosterols can all shift circadian clocks in the body and brain, including potentially in the SCN (reviewed in 46, 102, 115). Among these, caffeine is notable for its potent phase shifting effect in both mice and humans (25, 99). Most circadian nutritional studies have been completed in preclinical models and far less is known about diet composition impacts the clock in humans (but see 113). Furthermore, how different diet components contribute to the health impacts of time restricted eating has, to our knowledge, yet to be studied. Moreover, it will also be important to determine how food timing affects circadian rhythms in nutrient absorption which itself is rhythmic (56).

3.4. Food timing as countermeasure.

If flattened food intake across the day is a metabolic insult, then a high amplitude intake rhythm might be a rescue. Daily timed eating is a popular health promoting strategy, with a strong rooting in animal models (Figure 3c). For example, TRF can prevent diet induced obesity from high fat diets and can improve cardiovascular health (81). It is also an effective countermeasure against circadian disruptions by standardizing the food intake time and presumably improving either resonance or alignment of caloric intake with metabolic processes. TRF improves metabolic health in experiments of timed sleep restriction (15), simulated shift work (121), and chronic desynchrony and jetlag (101). This area of research has important implications for human health, for it suggests that particularly salient cues in the cyclic environment, in this case food, can prevent the deleterious effects of circadian disruption that might be caused by light cycles, activity/work cycles, or other cyclic biotic and abiotic cues.

3.5. Challenges in measuring food intake

Food intake is central to understanding unhealthy metabolism, yet reliably measuring intake patterns across the day is a challenge (7) (Table 1). The most common method is manual weighing of food at multiple times of day. Another simple method is video-recording: at the expense of caloric intake, a fine-grained behavioral record can be made without disturbing the animals. There are two main types of automated systems: pellet feeders and integrated mass sensors. Both are substantial investments though lab-built 3D printed pellet feeders have recently been developed (87).

Table 1.

Methods to measure food intake of laboratory animals

Method	Cost	Pros	Cons	Example Refs
Manual Weighing	Cost in staff time to be in lab at different times of day.	Accurate measure of food removed each period.	Coarse temporal resolution; animals are disturbed. Amount of spillage/grinding unknown.	(144)
Video	Time to score videos	Excellent behavior. Noninvasive	Amount ingested not known.	(109)
Pellet Feeders	Variable	Fine grain food seeking and pellet removal.	Pellet removal may not indicate intake. Cannot identify food caching.	(4, 87)
Mass sensors	Expensive	Fine grain food seeking. Accurate measure of food removal more likely correlated with consumption.	Sensor does not differentiate spillage or grinding from intake.	(30, 54)

Different restricted feeding methods are possible with all of the above methods. Food can be removed and provided manually each day, dropped automatically and removed manually, or food access can be controlled in commercial or lab-built systems with automated gates. Feeders that automatically open can also dictate consumption times, but if the food is readily removable, then the amounts in the feeder must be set with reference to baseline intake (49). Thus, overfeeding and underfeeding relative to “ad lib” is possible. Finally, a problem with all of these systems is non-ingestive food gnawing (grinding, shredding) that leaves crumbs or “ort” on the cage floor or in the feeder. Food grinding can lead to substantial differences in measured versus actual intake (~30% in some cases) (27). This is a clear issue of concern in studies that attempt to disentangle the relative contributions of calories consumed versus other metabolic mechanisms.

4. Circadian control of eating and metabolism in humans

Studies conducted in humans have built upon many of the aforementioned animal models in seeking to uncover potential underlying rhythms of hunger and metabolism in both real-world eating environments and tightly controlled conditions. Researchers have been curious about the impact of not only endogenous drivers of hunger and metabolism, but also how social and exogenous influences may impact the timing of food consumption and subsequent health. For centuries, many Western cultures have organized their eating patterns around three main meals a day (breakfast, lunch, and dinner) with snacking occasionally occurring between meals (42, 143). Though this organization of meal pattern is likely driven by work and social schedules, as animal models in both the laboratory and the wild do not exhibit similar temporal patterns, the timing of this caloric intake has recently come to the forefront of scientific interest as both a mechanism for poor health and a target for intervention. Here we will discuss endogenous rhythms for hunger and metabolism and how their timing may play a role in health and disease.

4.1. Benefit of human studies: separating activity, wakefulness, and eating.

Though light is the primary synchronizer of the central circadian clock in humans, exposure to exogenous stimuli may influence and/or mask endogenous rhythms. Thus, proper controls must be implemented to accurately assess these rhythms experimentally (Figure 4). In a constant routine protocol (93), volunteers are kept awake in constant conditions for an extended period to ensure that experimental outcomes are “unmasked” and primarily influenced by the endogenous circadian clock (43). Common constant conditions in human studies include: a dimly lit environment free of time cues to eliminate the alerting and phase shifting effects of light (26, 68); remaining sedentary in a constant posture to control for activity and postural changes on physiology (39); an ambient temperature within the thermoneutral range (22–24 °C) so as not to impact thermoregulatory physiology; hourly isocaloric meals to spread the intake of food across the 24h day and reduce the influence of feeding; and maintenance of wakefulness for one full circadian cycle to disassociate physiological impact of switching between sleep and wakefulness states (73, 131, 154). In this way, the typically covarying factors of wakefulness, eating, and activity can be disentangled in ways that would be difficult in laboratory animals.

Figure 4.

Examples of commonly employed sleep and circadian in-laboratory protocols to unmask endogenous rhythms in humans. Each protocol is displayed for a participant with an 8-hour habitual sleep duration beginning at 22:00. Gray bars denote dim-light settings (<5 lux of light) and black bars denote sleep opportunities (0 lux). (a) The constant routine protocol entails participants maintaining a constant posture with isocaloric meals provided in hourly increments to minimize potential masking of endogenous rhythms from activity or food intake. Constant routines must extend beyond 24-hours in order to capture the entire 24-hour rhythm without the masking of sleep. (b) The forced desynchrony protocol equally distributed activities and food timing across the 24-hour day by having the participants live on a non-24-hour daylength (20h in this example), thereby allowing the desynchronization of actives with circadian phase and sleep. (c) Acute reversal of sleep/wake, or slam-shift, protocol allows for testing if rhythms follow endogenous circadian rhythmicity or behaviors (i.e., eating) by testing for rhythmicity after a rapid inversion of the sleep/wake timing. Blue dots signify meal timing in each protocol.

Although a constant routine allows researchers to observe changes in circadian physiology, it also contains a major shortcoming in that a sleep debt accumulates throughout the protocol and resulting fluctuations in behavior cannot be distinguished as being circadian or an effect of sleep deprivation. To disassociate the influences of the sleep and circadian system, researcher Nathaniel Kleitman developed what is now known as a forced desynchrony (FD) protocol in Mammoth Cave (69), presaging the T-cycle experiments in plants and animals that would follow. In a FD protocol, participants live on a compressed or dilated “day” (e.g., 20h or 28h long, cycles to which the circadian system cannot entrain) (36). This allows the participant to sleep or eat on these short or long schedules and enables the researcher to observe physiologic and behavioral outcomes across the entire endogenous circadian period (τ~24.1h in humans) with minimal accumulation of sleep debt, thus disassociating the effects of the sleep and circadian system (36).

A third protocol employed to study circadian rhythms is an acute reversal of sleep and wakefulness (i.e., “slam-shift” schedule). While adhering to most constant routine conditions, researchers have participants sleep during their biological day and perform wakefulness activities during their biological night (90, 152) . If the regular cycling of biological or behavioral outcomes persists regardless of wakefulness or sleep states, then these outcomes are considered circadian. If these biological or behavioral rhythms invert, the outcomes are influenced more by sleep and wakefulness states (19, 40). The reversal of sleep and wakefulness protocol enables researchers to observe acute relationships between the sleep and circadian system and is a useful tool for examining populations that must rapidly shift their schedules (e.g., night shift workers).

Taken together, these in-laboratory protocols can allow for the investigation of 1) underlying endogenous rhythms of an individual, 2) how behaviors influence physiology independent of the endogenous rhythms, and 3) the combination of the rhythm with the behavior on outcomes.

4.2. Lessons from the lab. Mechanisms and functional dissection of clock contribution

Before the use of the aforementioned laboratory paradigms, early-work to examine human behavior across the 24h-day would include participants living in an environment free of time cues and choosing their sleep, wake, and eating times simply whenever they felt the drive to perform those behaviors. In seminal studies performed in underground isolation units, Aschoff and colleagues (11) found that when allowed to self-select their meal times without time cues, the majority of participants (83%) ate 3 meals a day, however the timing of these meals either lengthened or compressed depending on the length of the participants’ wakefulness episode which differed depending on circadian phase. While it is difficult to tease apart any potential socially driven influences to eat 3 meals a day, or the linear build-up of hunger across wakefulness, these data do suggest that there is an endogenous drive for hunger at differing times of day. Indeed, utilizing forced desynchrony protocols, and thereby separating the circadian influence from behavior, peak and trough hunger ratings on a visual analog scale are reached at ~20:00 and ~07:50, respectively, with the peak potentially driving caloric intake in preparation for a nightly fast and the trough delaying hunger in the morning to prolong the rest period (124). There is also an entrainable circadian anticipatory hunger rhythm, such that if meals are provided at a consistent time for six days and then later restricted during a constant routine protocol in an environment free of time cues, participants will continue to report higher levels of hunger at the time that the meals would have been provided (59). Moreover, the circadian drive for hunger is supported by objective eating patterns such that when individuals are sleep restricted and food is available ad libitum, participants eat more in the evening hours as compared to any other time during the day (83). Interestingly, the circadian pattern of hunger can be shifted based on when food is provided. For example, providing a higher proportion of daily calories in the morning hours can result in lower hunger levels throughout the day (119) and providing a higher proportion of daily calories in the evening increases hunger while also lowering daytime energy expenditure in individuals with overweight/obesity (148). It remains unclear, however, what mechanism is driving increases in hunger during the circadian evening or in anticipation of meals, or how sleep loss may or may not play a role in enhancing hunger at certain times of day.

Regarding energy metabolism, energy expenditure also follows a 24-h rhythm with different aspects of metabolism exhibiting different rhythms. Under conditions of a 34h constant routine protocol, Krauchi and Wirz-Justice (73) observed a 24-h rhythm in heat production with a peak in the middle of the day and a trough between midnight and 06:00. Spengler and colleagues similarly found that CO₂ production, a byproduct of energy metabolism, also exhibited circadian rhythmicity with a peak at ~18:00 during a 41-h constant routine protocol (132). Rynders and colleagues (120) however, did not find significant circadian rhythmicity in energy expenditure as measured via indirect calorimetry in a 26-hour constant routine, potentially due to an observed low amplitude of energy expenditure fluctuation that is being masked by isocaloric food intake every hour. Using a forced desynchrony, and thereby better able to separate and unmask the impact of fasting energy expenditure vs a continually fed state imposed by a constant routine, Zitting et al. (163) were able to identify significant circadian rhythmicity in fasting energy expenditure using indirect calorimetry, with a peak occurring in the evening hours. McHill et al. (92) extended these forced desynchrony findings to demonstrate that the circadian rhythm in energy expenditure persists under conditions of mild-to-moderate exercise intensity and that obesity may influence the timing of this rhythm by delaying the minimum of energy expenditure.

It is important to note that the magnitude of energy expenditure fluctuation across the 24-day is minimal, and thus easily masked, making the implications for the interaction between caloric intake and resting energy expenditure depending on circadian phase unclear. Slam shift experiments suggest that the thermic effect of feeding, or the energetic response to consuming food, is decreased during the circadian evening as compared to daytime hours (90, 96), however this has been postulated to be attributed to the underlying rhythm in resting energy expenditure (118). Though use of tightly controlled in-laboratory procedures are necessary to understand mechanistic connections between food intake and the circadian rhythm in metabolism, these protocols do not generalize well to the real-world settings. Further, the majority of controlled circadian protocols have been conducted with young healthy volunteers. Therefore, new methods to bridge lab and field are still needed to better understand circadian rhythms in human health.

4.3. Lessons from the field. Epidemiology and natural experiments

In the modern 24-hour society, habitual schedules include erratic eating patterns with individuals consuming meals at all times of day across all circadian phases (48, 89). This is further complicated by the >20% of the working population that work overnight shift work hours, which pushes food intake into the nighttime hours (72, 110). However, using these data from field-based studies, we can glean how eating patterns are associated with more advanced disease progression. Furthermore, technological advances such as photographic food diaries have enabled researchers to gain further insights into human eating behaviors while living on their habitual schedules.

One of the most common observations of food timing with disease has been the interaction between late meal timing and obesity (Figure 5). Individuals working overnight shift work, a drastic example of eating the majority of calories during the night, have up to a 35% increased risk for abdominal obesity (140). Moreover, patients with Night Eating Syndrome, a condition in which persons follow a pattern of consuming >50% of their daily calories after 19:00, have a higher body mass index (BMI) as compared to those without Night Eating Syndrome (33). In less extreme examples, the distribution of meal timing has been associated with differences in body composition. Those that skip breakfast tend to push a greater amount of their daily calories to a later clock time (66) and tend to have more difficulty losing weight in diet programs (61). Indeed, numerous reports have found that those eating a greater proportion of their calories later in the day have a higher BMI and body fat percentage (17, 91). The reverse of this association is also true: those that eat earlier have lower obesity risk and can more easily lose weight during weight-loss programs (47). Thus, this physiological relationship between the circadian timing of food intake and body weight may be leveraged to improve health. Using time-restricted eating (TRE) protocols, wherein a person’s eating duration is shortened to 6–8 hours daily, individuals have been able to reduce weight and hunger (48), improve oxidative stress (141) lower blood pressure (155), improve blood lipids (8), and better manage glucose tolerance (100). Excitingly, TRE regimens have been found to be as effective as caloric restriction diets (79). Mechanistically, while increasing the daily fasting duration may be one way in which TRE protocols work (107), TRE protocols are most effective if the eating duration window is maintained in the morning hours (157), matching the in-laboratory work that suggests food intake in the evening would occur at a circadian time when the thermic effect of food is at its lowest (90). More work is needed in this area to pinpoint exact mechanisms of action, such as the interaction between schedule and circadian phase, and to determine the most effective eating schedules that are not only the most successful in improving health outcomes, but also feasible for individuals to maintain daily.

Figure 5.

The timing of food intake and cardiometabolic health. Eating during the nighttime hours, thereby creating misalignment between the central circadian clock and behavior, is associated with increased percent body fat composition, weight, oxidative stress, and glucose intolerance, amongst other adverse outcomes, as compared to those that eat earlier in the day. Restricting the timing of food intake to the daytime hours, such as with time-restricted eating protocols, results in improvements in a number of health outcomes and may be a feasible therapy for those with disease or to prevent disease.

Field studies have strengthened our understanding of how an individuals’ food timing in real-world scenarios may interact with health and they complement the in-laboratory work on mechanistic pathways. However, there are challenges in fieldwork that should be considered when evaluating study findings. For example, just as for the laboratory animals, dietary patterns are often difficult to accurately capture, and methods such as food frequency questionnaires or 24-hour dietary recall can require high researcher burden (139), differ in their outputs thereby decreasing reliability (142), and for the purposes of this review, do not allow real-time tracking of the timing of food intake (50). Thus, methods utilizing smartphones have been developed to better capture the timing, content, and amount of food intake (117). These methods may also improve compliance as taking a picture is simpler than needing to recall everything that was consumed. Photo-based records, however, are not without limitation; it is often difficult to accurately estimate portion size, particularly if the picture-taking method is not uniform. Other challenges of fieldwork may include participant reliability, confounding influences on eating behaviors (e.g., holidays, social gatherings, food availability) that may not accurately reflect habitual behaviors, and not having markers of internal physiology that could impact food timing, although proximate markers such as morning/evening preference (63) and mathematical modeling (88) have promise as estimates of circadian timing for eating studies.

5. Translational pitfalls and potential

5.1. Effects of body size in comparative metabolic studies

Poor metabolism is a common feature when internal clocks are disrupted or when meal times are shifted in both humans and rodents, strongly suggesting common mechanistic pathways. This raises the exciting potential for bench to bedside translational research. Nevertheless, some caution may be needed in interpretation given species differences and in particular, body size.

Comparative physiology rests on the maxim that a tractable physiological model exists that can recapitulate a human system (75). Body size is a constant challenge in this. The problem of metabolism and scaling preoccupied physiologists for decades, illustrated by humorous (and serious) discussions of whether the Lilliputians calculated the food needs of Gulliver correctly (95). Body size influences mass-specific metabolic rate, daily nutrient needs, and the relative peril of long fasting intervals. Humans can survive weeks without food while small-bodied species succumb to starvation far sooner. Reactions to fasting will differ in terms of rate of depletion of internal stores such as glycogen, the incorporation of body temperature adaptations such as torpor, and how quickly circulating nutrient and glycemic signals change as the body moves towards a catabolic state.

For circadian timing, body size affects the rate at which nutrient signals wax and wane over the day as well as the magnitude of their response to eating, fasting, and refeeding. Mice eat ~3/4 of their total intake during the night in many small meals. The amount eaten also depends on strain and on temperature (64). Humans on the other hand tend to eat close to 100% of their intake between wake time and bed time. The difference in meal structure leads to very different circulating glucose concentrations (Figure 6). Continuous glucose monitoring now provides a method to easily track daily changes in glycemic control. In mice, numerous small meals lead to a fairly constant glucose concentration with small increases around lights-on and lights-off when food intake is maximal (116). In humans, glucose values vary widely in comparison, with strong postprandial excursions (62).

Figure 6.

Continuous glucose monitoring from ad libitum fed and fasting mice (left) and from humans eating three identical meals (M) across 12h or 6h (right). Shading indicates lights-off: the active phase for the mice and the sleeping time for the human participants. Data on the left kindly provide by Dr. Alex Banks (116). Data on the right adapted from (62); distributed under a CC BY 4.0 license. .

Unsurprisingly, mice and humans have dramatically different responses to short fasts. Studies of restricted feeding in mice typically employ fasts of 12–16 h, similar to fasting times in human TRE. In humans, morning glucose concentration is the same after both 12 and 18h fasts. In the latter, glucose stabilizes at its basal level much earlier (Figure 6). In fasting mice, however, circulating glucose progressively declines over 12 h. This large decline is accompanied by significant hypoinsulinemia within 2h of the fast beginning (97); in contrast, insulin stabilizes at a basal level overnight in humans (114). Additionally, fasts of this duration cause 10–15% drops in body weight in mice (12, 64): hepatic glycogen is completely depleted and mice enter a strong catabolic state in order to protect against starvation. This is a far cry from the modest changes that humans experience overnight. It means that in many cases, the comparative model mouse is being studied under comparable fasting durations but absolutely uncomparable metabolic states.

To address this, recent recommendations are to reduce fasting intervals in animal studies of fasting metabolic parameters. Intervals of 6 h or even 2 h may be preferable for measures of glycemic control and insulin signaling (28). In studies of circadian rhythms, similar care may be needed to determine physiologically relevant mechanisms that are shared between rodents and humans when food timing is restricted.

5.2. Physiological and translational relevance

We need not be pessimistic in translation, however! First, common mechanisms across body sizes may indeed be common though sometimes obscured by imposed ingestion patterns. Second, the greater metabolic stress and greater responsiveness of small animals may reveal non-physiological mechanisms that can nevertheless be harnessed pharmacologically as treatments in humans.

To investigate the first, we studied mice with natural eating patterns to avoid the long fasting interval typical of TRF. Surprisingly, we found that natural eating patterns were at best a weak zeitgeber for peripheral organs in mice (156). We used pellet feeders to mimic the natural intake pattern, with approximately 25% of intake during the light phase. Shifting the intake pattern by 12h led to only small phase shifts of the peripheral organs, far less than the ~12h shift anticipated based on TRF studies (37, 138).

These studies indicate that the established model of circadian orchestration of body clocks via food intake (126) may have limited relevance during conditions of ad lib feeding for a small animal. Instead, the SCN retains an important synchronizing role that is hidden during restricted feeding schedules. This is still encouraging for translational relevance. These smaller food-related effects in ad libitum conditions qualitatively match the small clock gene phase shifts that have been reported in response to shifted meal timing in humans (151). Further, food is not always a weak zeitgeber for humans. A week of altered meal schedules can entrain circulating glucose rhythms and hunger (59), and simulated shift work can cause metabolomic desynchrony with some metabolites following the imposed wakefulness/meal schedule and others remaining aligned with the SCN’s clock (130). Therefore, both the strong resetting of TRF in rodents and the weak resetting observed in ad libitum fed rodents may reveal physiologically relevant mechanisms. And even if not physiologically relevant, TRF models offer the potential for identifying and studying potential chronotherapeutic mechanisms.

6. Conclusions and Future Directions.

“Let food be thy medicine and medicine be thy food.”

– Hippocrates

The timing of food intake is increasingly being recognized as a risk factor for poorer cardiometabolic health in both animals and humans. The endogenous circadian clock has evolved to promote intake at optimal times when an organism is intended to be awake and active, however, with the invention of electrical lighting, humans have created an environment in which whenever their eyes are open, their mouths are also typically open and eating. Within this review, we have provided a critical appraisal of the significant literature pertaining to the effects of eating time on health and how resonance of food timing, light-cycle timing, behavior, and internal clocks are integrated in health and disease. Animal models, particularly with the ability to explore mutations in circadian genetics, have allowed us to further our mechanistic understanding into how food timing can either synchronize or disrupt the endogenous clock and subsequent behaviors. With some caveats, these findings may be translated into humans; an added benefit of this bench-to-bedside research is that human-subjects experiments give the added ability to tease apart behaviors from endogenous physiology utilizing tightly controlled in-laboratory designs. Furthermore, work in the field and technological advances of wearables and photo diaries have allowed for additional understanding into how eating behaviors in real-world scenarios are associated with disease. Optimistically, clinicians and public health campaigns can use the timing of food intake as a feasible means to improve many aspects of health, though further work is needed in understanding the mechanisms behind TRE and the interplay between the restriction of food timing and the circadian system in humans. The timing of food intake and the circadian system are intricately intertwined, and the importance of “being on time” with our eating habits is vital for optimal health.

Terms and Definitions

Circadian

Endogenous rhythm with a free-running period close to 24h

Diurnal/Nocturnal

Day-active / night-active

Zeitgeber

“Time giver” or an entraining cue

Entrainment

The synchronization of circadian rhythms to zeitgebers

Period

The amount of time it takes for a rhythm to complete one cycle

Phase

A measure of the timing of a rhythm. The time at which a particular event in the cycle (e.g. peak or trough) occurs with respect to some external timing reference point, such as lights-on or –off

Phase Advance

A phase shift to an earlier time

Phase Delay

A phase shift to a later time

Coupling

Two or more rhythms that share a common period but may differ in their phase

Internal misalignment

A state characterized by internal rhythms that have an incorrect phase relationship

External misalignment

A state characterized by internal rhythms that have become desynchronized from the environment. The misalignment can be stable, when both have periods of 24h, or transient when rhythms precess with respect to the external driver

Shift work

Work that occurs outside typical working hours. Many schedules fall under shift work: night, swing, graveyard, early-morning, shoulder, rotating, etc. And not all of these occur at the same time, which makes quantifying shift work in analyses difficult. In many epidemiological studies, all non-standard shifts are grouped together as “shift work.”

Social jetlag

Difference between midsleep time on work/school- and work/school-free days. It is a useful measure of how much internal phase may be advancing and delaying each week

T-cycle

The length of “day” in experiments, for example a 20h day of 10h light and 10h dark for rodents or a 28h day with 18h40m wake and 9h20m sleep for humans (see forced desynchrony in section 4.1)

Abbreviations

AgRP

Agouti-related protein

CRY

Protein product of Cryptochrome genes

DMH

dorsomedial nucleus of the hypothalamus

FD

forced desynchrony

PER

Protein product of Period genes

POMC

Proopiomelanocortin

PPAR

peroxisome proliferator-activated receptor

SCN

Suprachiasmatic nucleus

TRE

Time restricted eating

TRF

Time restricted feeding