Human glucose rhythms and subjective hunger anticipate meal timing

consumed

Correspondence j.johnston@surrey.ac.uk In brief Isherwood et al. compare the effect of regular large-versus small-meal schedules in humans using highresolution glucose monitoring and hunger scores. Glucose concentration lowers in anticipation of regular large afternoon meals. Differences in circadian rhythms occur even after the preceding largeversus small-meal patterns have stopped.

RESULTS
Circadian rhythms regulate many key aspects of physiology, including metabolism. [4][5][6] Mammals possess clocks throughout the body that together comprise the circadian timing system. A central clock in the suprachiasmatic nuclei (SCN) of the anterior hypothalamus is entrained by environmental light and synchronizes ''peripheral'' clocks in the brain and body through a range of mechanisms. These mechanisms include neuronal pathways, endocrine secretion, core body temperature, and behaviors such as sleep-wake and feeding-fasting cycles. 7,8 The timing of food intake has been shown to be a dominant synchronizer of rodent peripheral clocks, including in the liver and adipose tissue. 9,10 Animal studies over many decades have also revealed anticipation of food availability, linked to a food-entrainable oscillator, 3,11 the locations of which remain unclear. In humans, we have previously demonstrated that circadian rhythms of glucose homeostasis are synchronized to a pattern of 3 meals per day. 2 Despite this, how meal size influences circadian rhythms is poorly understood. It is also unknown whether the human circadian system predicts meal timing under conditions of restricted food availability when food is only available for short periods of the day.
We tested the hypothesis that the human circadian system can anticipate large meals. Young, healthy male participants, n = 24, undertook a 9-day residential laboratory protocol (Figure 1). During days 1-6, under entrained conditions, half the participants received 14 identical hourly small meals (small meal group), whereas the other half received 2 identical large meals, 7.5 and 14.5 h after waking (large meal group). Details of the methods including the food are provided in STAR Methods. This meal timing pattern resulted in the small meal group initiating food consumption 7 h before the large meal group, with both groups having their last meal at the same time of day. The meal pattern also resulted in both groups having consumed the same cumulative energy at the time of day when the large meal group had consumed each of their meals. Following the 6 days of entrained conditions, all participants then underwent a 37-h constant routine protocol, which enables measurement of endogenous circadian rhythms by removing daily variation imposed by overt rhythms in the environment and behavior. During the constant routine, hourly saliva samples were collected to measure the timing of the dim light melatonin onset (DLMO), the gold standard marker of the SCN clock phase. Throughout the protocol, interstitial glucose concentration was monitored by continuous glucose monitoring (CGM) devices with a 15-min resolution. Subjective hunger was assessed hourly during the waking periods of days 2, 4, and 6, plus hourly throughout the constant routine.
We first examined the effect of our small vs. large meal patterns on melatonin rhythms. We have previously shown plasma melatonin rhythms to be unchanged after a 5-h delay in meals in a fixed light-dark cycle. 2 Consistent with our previous work, there was no difference in melatonin profiles between the small and large meal groups in this study ( Figure 2). DLMO was 22.68 ± 0.3 and 22.5 ± 0.17 h in the small and large meal groups, respectively (p > 0.05).
The CGM data enabled analysis of high-resolution glucose dynamics over the 6-day period in a highly controlled environmental and behavioral setting (entrained conditions), followed by the 37-h constant routine, which reveals endogenous glucose rhythmicity. Over the course of the entrained laboratory protocol (days 1-6), there was no difference in average glucose concentration (5.48 ± 0.11 vs. 5.51 ± 0.07 mmol/L; p > 0.05) between the two groups. However, the daily profiles of glucose concentration exhibited clear differences between groups ( Figure 3).
During the entrained conditions in the laboratory (days 1-6), interstitial glucose concentration time courses in the small meal group increased from around the onset of wake and remained elevated at a similar level during the feeding/light period. Glucose concentrations then rapidly declined after the last small meal and remained low during the fasting/dark period. In the large meal group, there was a transient increase in glucose concentration around the onset of wake followed by a gradual decline in glucose concentration until the first of the 2 large meals ( Figure 3). There was a large increase in glucose concentration after each of the 2 meals, with the area under the curve of the postprandial excursion being significantly higher after the second meal (324 ± 32 vs. 487 ± 54 mmol/L.min; p < 0.01). There was also a significant effect of study day (p < 0.01), with post hoc tests revealing a significant difference between the first vs. second meal on days 1-4 and day 6 (p < 0.05) but not on day 5 (p = 0.1).
At the start of the constant routine, there was an initial increase of glucose concentration over the first 2-3 h in both groups. In the group that had previously consumed hourly small meals, glucose concentration continued to rise gradually over the subsequent 10 h. In the group that had previously consumed two large meals, the rise in glucose concentration over the initial 2-3 h of the constant routine was followed by a rapid decline over the following 3-4 h, resulting in a lower glucose concentration than the small meal group during the time of eating in the preceding entrained protocol. Both groups then exhibited circadian rhythmicity of glucose concentration ( Figure 4), with data from the 2 groups in approximate antiphase to one another (acrophase À6.13 ± 0.64 vs. 6.28 ± 1.01 h, small vs. large meal groups respectively; p < 0.001). The amplitude was also significantly higher in the large meal group (0.19 ± 0.03 vs. 0.32 ± 0.06 mmol/L small vs. large meal groups respectively; p < 0.05). The participant entered the laboratory on the afternoon of day 0. On days 1-6, they slept in the dark (0 lux), in individual rooms from approximately 23:00 to 07:00 (dark gray shading). They were either fed two identical isocaloric meals each day at 7.5 and 14.5 h after wake-up ($14:30 and $21:30 h), in the large meal (L) protocol (shown in the top panel) or 14 hourly meals, each containing 1/14 of their daily calorie requirements in the small meals (S) protocol (shown in the bottom panel). On days 7 and 8, they were under constant routine conditions (gray shading) (<8 lux, remained awake, semi-recumbent position and consumed hourly isocaloric snacks). Interstitial glucose monitors were inserted 3 days before entry to the laboratory and remained in place until the end of the study. Participants were discharged on day 9. See also Data S1.

OPEN ACCESS
Subjective hunger data, collected on days 2, 4, and 6 and across the constant routine, are presented in Figure 3. On days 2, 4, and 6, there were significant differences between hunger scores across each day (p < 0.001) and a significant interaction between time and meal group (p < 0.001), without an overall effect of the meal group (p = 0.16). In the large meal group, hunger exhibited a linear increase from wake-up through to the first meal, declined greatly after the meal, and then showed another linear increase up to the second meal. As expected, there was much less daily hunger variability on days 2, 4, and 6 in the small meal group. In the constant routine, both groups showed a similar overall effect of time (p < 0.01), with reduced hunger in the biological morning, as reported previously. 2,12 There was an overall effect of group (p < 0.05) but no time 3 group interaction (p = 0.99). In the large meal group, there was an increase in hunger preceding projected mealtime in the constant routine. Immediately after the time of the anticipated first large meal, there was a sharp decline in hunger on both occasions that this occurred in the constant routine.

DISCUSSION
These high-resolution CGM and hunger data, collected within highly controlled laboratory conditions, greatly increase our understanding of the effects of meal timing on human biology under entrained conditions and under constant routine. In the large meal group, during entrained conditions in days 1-6, an initial rise of glucose during the early morning sharply declined over the following hours in anticipation of the first large meal. These data suggest that human physiology entrains to, and can predict, the timing of regular and temporally restricted food availability. In the constant routine section of the protocol, the two groups exhibited anti-phasic circadian rhythms of glucose concentration, despite there only being a 7-h difference in the onset of daily food consumption in the previous 6 days and no difference in the timing of the last daily food consumption. Furthermore, in the constant routine of the large meal group, subjective hunger scores increased before anticipated mealtimes and declined sharply after the projected time of the first meal. Together, these findings support the existence of food anticipation in humans.
To our knowledge, this is the first study to investigate the possibility that humans anticipate large meals, following a period of restricted food availability. There was no difference in average daily glucose concentration between the two groups indicating that, in these healthy young men, the dietary intervention did not acutely disrupt overall glucose homeostasis. Despite this, the dynamics of glucose concentration varied greatly between groups.
During days 1-6 of the study, glucose concentrations in the small meal group followed a simple pattern: glucose concentrations increasing in the early morning to reach a plateau that was maintained throughout the day, before declining to nocturnal baseline fasting values. The large meal group exhibited a declining pre-meal glucose concentration, followed by large postprandial increases. The glucose response was prolonged after the second (evening) meal, compared with the first (afternoon) meal, despite the meals being identical in size and composition. The elevated postprandial response to the second meal may be explained by the well-reported daily variation in glucose tolerance. 6,13 Our data from days 1-6 also suggest the presence of physiological anticipation of the first large meal. As observed in the small meal group, data from the large meal group also revealed an increase in glucose concentration in the early morning. However, glucose concentration in the large meal group then declined in the 3-4 h prior to the first large meal. This pre-prandial change may be analogous to the food anticipatory responses that are well known in animal models. 3,11 When food availability is restricted to a fixed daily pattern of only a few hours per day, many animals exhibit marked increases in food anticipatory activity (FAA) that persists when fasted, even in SCN-lesioned animals. Mistimed food intake (during the day in rodents) ''uncouples'' timing of peripheral clocks from the SCN, 9 and the resulting conflict in timing between light and food associates with increased adiposity and impaired glucose tolerance. 14 Less is known about the effects of food timing on circulating glucose levels in rodents, which typically eat throughout their active phase rather than in delineated meals, but there are reports that restricting feeding to specific parts of the day affects the phase and amplitude of glucose rhythms. [15][16][17][18] However, until now, physiological anticipation of meal timing has not been thoroughly explored in humans.
The constant routine portion of our protocol (days 7 and 8) is a well-validated method of unmasking endogenously driven human circadian rhythms. 19 Throughout the 37-h constant routine, participants were kept awake in constant dim light, with constant posture and feeding of identical hourly snacks irrespective of the meal patterns they received during days 1-6. This removes overt cycles of light and dark, sleep and wake, and feeding and fasting. In this study, both groups exhibited an initial rise of glucose concentration at the start of the constant routine. This initial rise was followed by a sharp drop in glucose levels in the group previously fed large meals but a continued rise in the group previously fed small meals. This contrast between groups associated with the emergence of antiphasic glucose rhythms with glucose peaking in the subjective day for the small meal group but in the subjective night for the large meal group. We have previously reported that a 5-h delay in provision of a standardized 3-meal pattern causes a 5-to 6-h delay in the circadian phase of plasma glucose during constant routine conditions. 2 In this study, the antiphasic (c. 12 h) difference in the glucose rhythm phase occurred despite the meal pattern in days 1-6 having only a 7-h difference in the morning onset of energy consumption and the last energy consumption occurring at the same subjective time of day in both groups. Our data therefore may indicate that the size, as well as phase, of energy intake is important for synchronization of circadian glucose homeostasis. Future work will be necessary to specifically test the importance of meal size. Consistent with our previous human data and studies in rodents, the SCN phase as measured by saliva melatonin was not affected by meal timing under conditions of energy balance. This strongly suggests that the changes we report in glucose and appetite are driven by one or more peripheral clocks, outside of the SCN. By delaying mealtimes by 5 h within a fixed light-dark schedule, we previously showed that human plasma melatonin rhythms remain fixed to the ambient light-dark cycle. 2 In rodents, peripheral clocks preferentially synchronize to food timing; by contrast, the SCN clock remains synchronized to the light-dark cycle over feeding times, unless in a situation of hypocaloric energy consumption or food with a high hedonic value. 20 In this study, each participant's daily energy intake was adapted to meet their energy requirements, determined using the Mifflin St Jeor equation. 21 Whether the human SCN clock will synchronize to mealtimes in a state of negative energy balance remains unknown.
Our study includes certain limitations. On medical advice, as explained in the screening section of STAR Methods, we only recruited male participants for the study. We have not directly measured plasma or serum glucose concentrations here. However, previous validation of CGM showing that it accurately reflects blood glucose concentration, 22 together with the highresolution data that CGM provides, makes it a powerful research tool. In our study, participants in the large meal group had restricted food availability at times that were within the biological wake/active phase. In contrast, many animal studies of FAA employ ''mistimed'' food availability, restricted to the usual sleep/rest phase. Despite the potential ethical and sleep deprivation issues, it is likely that a human protocol in which food availability is restricted further away from midday would yield even clearer anticipatory responses. Finally, future work will be needed to elucidate the mechanisms underlying the temporal changes reported here. Key candidate molecules include cortisol, a glucocorticoid hormone that exhibits strong rhythmicity and regulates plasma glucose, metabolism, and appetite. Nonetheless, we have now provided evidence of human food anticipation that could be used to design appropriate future human studies.
In conclusion, we report the first evidence that humans exhibit physiological food anticipation of restricted food availability occurring at the same time of day. Our data also suggest that synchronization of circadian glucose rhythms may be driven not only by meal timing but also by meal size. Together, these findings increase our understanding of circadian metabolism and chrono-nutrition in humans.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:  Data show mean ± SEM (n = 11 small meal group, black; n = 9 large meal group, red).

Materials availability
This study did not generate new unique reagents.

Data and code availability
All data reported in this paper will be shared by the lead contact upon request. This paper does not report original code. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
Twenty-four healthy male volunteers were studied. Study participants were recruited by the Surrey Clinical Research Facility (CRF). Only male volunteers were considered based on risk assessment and the advice of the CRF medical team. The overall study included extremely high frequency of blood sampling during the constant routine (60 blood draws: every 30 minutes for 30 sequential hours). Previous studies have demonstrated the need for frequent re-cannulation and several phlebotomy attempts due to difficulty with the vein integrity of female participants with frequent blood draws. This has been associated with higher risk of relevant side effects of cannulation/phlebotomy, which could have compromised the participants' well-being. After an initial telephone screen to check they met the broad inclusion/exclusion criteria, questionnaires for medical history, sleep habits and chronotype and diet history ensured the participants were healthy and had a regular sleep schedule and were not extreme morning or evening chronotypes.
Inclusion criteria included being male, aged between 18 and 35 years and BMI between 18 and 30 kg/m 2 inclusive at screening. Habitual (at least 5 days per week) hours in bed per night between 7-9 hours which included a bedtime between 22:00-01:00 and waketime between 06:00-09:00. ESS score <9 indicative of normal range daytime sleepiness, HÖ (diurnal preference) score between 30-70 (normal range) and PSQI score <5 indicative of satisfactory sleep quality and sleep patterns.
Shift workers involved in night work within the past six months and volunteers that had travelled across more than two time zones within a month before the study were excluded. REAGENT

Report METHOD DETAILS
Pre-laboratory protocol A 10-day pre-laboratory routine, a previously validated standard for our human chronobiology experiments 2,23,24 was designed to control the participants' circadian phase of entrainment and minimise sleep deprivation before they entered the laboratory. The bedtimes, based on their habitual sleep-wake pattern, included a strict 8 h in-bed window without deviation from their in-bed and get-up times by more than 10 minutes each night. Participants were instructed to remain in the dark, trying to sleep, without distraction (such as no light emitting devices) which was verified by calling the study voicemail morning and evening and wearing Actiwatches (Light-actiwatches [Cambridge Neurotechnology, Cambridge, UK]) that measure motion and light exposure. They could nap within a 4 h window in the afternoon (centred 12 hours away from the midpoint of their night-time sleep) if they chose to do so. Naps scheduled this way are thought not to phase shift circadian rhythms. 25 They had to obtain 15 minutes of outdoor natural light (without wearing sunglasses) within 90 minutes of waking up to strengthen the synchronisation of their circadian rhythms. 26,27 They also had to consume less than 100 mg caffeine within the first 3 hours after waking and up to 4 units of alcohol per day.
Three days before the laboratory study the actigraphy and diet diaries were checked for compliance, and they had a continuous glucose monitor (CGM) (Freestyle Libre 2, Abbott Laboratories Limited) inserted into their non-dominant arm. During the last 3 days of the pre-laboratory routine, participants were instructed to refrain from strenuous exercise, by remaining within the 'rest', 'really easy' and 'easy' portion of the Borg CR10 scale, 28 which is part of our standard pre-laboratory routine. They were also instructed to refrain from consuming alcohol and caffeine, known confounders of circadian rhythms and they had to eat their meals within specific time windows; breakfast; 08:20 h to 09:50 h, lunch; 13:45 h to 15:15 h and dinner; 20:00 h to 21:30 h to reflect the meal timing windows in the laboratory protocol.

Monitoring compliance
To confirm compliance to the pre-laboratory routine, the participants were instructed to call the study voicemail within 10 min before going to bed and 10 min after waking and wear two light and motion recorders, which record physical activity and environmental light levels. One Actiwatch was worn around the neck to monitor light exposure as close to eye level as possible and the other was worn on the non-dominant wrist to record activity. They had to complete a sleep diary, recording bed and wake times, estimates of sleep onset latency and awakenings, sleep quality, whether they removed the actiwatches and for how long and time of their 15-minute exposure to outdoor natural light. They had to complete a food and physical activity diary, recording when, what and how much they ate and drank, and when and how long they exercised and what type of exercise they did. The CGM confirmed compliance to the meal eating times.

Laboratory protocol
The 24 participants were studied in a parallel design. Each laboratory session contained only one protocol (small meals or large meals) so that all the participants were following a similar schedule in the communal lounge. Each protocol (small meals or large meals) was run alternately in the facility. The Participants were allocated to 'small meals' or 'large meals' groups on a rolling basis as they passed screening. Participants were supervised throughout the laboratory sessions by medical/clinical research staff.
Upon entering the CRF in the afternoon of Day 0, the participants were assessed to confirm compliance to the pre-laboratory routine and a review of medication, an alcohol breath test and a urine sample for analysis of cotinine and drugs of abuse were performed.

Laboratory environmental conditions
During the meals schedule, lighting was set at approximately 500 lux (direction of gaze) throughout the wake periods (07:00-23:00 h) and 0 lux during sleep opportunities (23:00-07:00 h). During the CR the lighting was set to < 8 lux, as used in our previous studies. 2,23 During wake periods of the 6-day meal schedule participants had knowledge of clock time, were able to move around and were predominantly seated in the communal lounge. They were allowed to read, watch TV/films, access the internet and play computer or board games and visit the bathroom but were not permitted an activity that required physical exertion or excessive postural change.

Meals
During Days 1-6, the participants consumed either 14 isocaloric meals (small meals) or 2 x large meals, each meal containing 50% food energy requirements. The large meals were timed at 7.5 hours and 14.5 hours after each participant's wake-up time. The first small meal was given to the participant 90 minutes after waking. The macronutrient content of the meals was: 55% carbohydrate, of which 15% sugars; 15% protein; 30% fat, of which 11% saturated fat, as in our previous research. 2 Meals were eaten in isolation to eliminate the impact of the smell and sight of other participant's food on metabolism, appetite and mood. Meal size was adapted to each participant's daily energy requirement, calculated using Mifflin St Jeor equation. 21 A daily allowance of 1.5 litres of water were given to the participants throughout the laboratory session during wake periods. If the participants required more water, it was given and recorded. Within 30 minutes of wake, after the participant had emptied their bladder, they were weighed to monitor changes in weight throughout the laboratory session. Body weight was maintained within 5% of body weight at the start of the study in every participant (Data S1). Over the course of the study, participants lost on average 1.75% of their body weight but there was no significant overall effect of meal group (p = 0.107).