Chronic circadian disruption on a high-fat diet impairs glucose tolerance

Abstract

Background:

Nearly 14% of Americans experience chronic circadian disruption due to shift work, increasing their risk of obesity, diabetes, and other cardiometabolic disorders. These disorders are also exacerbated by modern eating habits such as frequent snacking and consumption of high-fat foods.

Methods:

We investigated the effects of recurrent circadian disruption (RCD) on glucose metabolism in C57BL/6 mice and in human subjects exposed to non-24-h light-dark (LD) schedules vs. those on standard 24-h LD schedules. These LD schedules were designed to cause circadian misalignment between behaviors including rest/activity and fasting/feeding with the output of the near-24-h circadian system, while minimizing sleep loss, and were maintained for 12 weeks in mice and 3 weeks in humans. We examined interactions of these circadian-disrupted schedules compared to control 24-h schedules with a lower-fat diet (LFD, 13% in mouse and 25–27% in humans) and high-fat diet (HFD, 45% in mouse and 45–50% in humans). We also used young vs. old mice to determine whether they would respond differently to RCD.

Results:

When combined with a HFD, we found that RCD caused significant weight gain in mice and significantly impaired glucose tolerance and insulin sensitivity in both mice and humans, but this did not occur when RCD was combined with a LFD. This effect was similar in both young and older mice.

Conclusion:

These results suggest that reducing dietary fat may protect against the metabolic consequences of a lifestyle (such as shift work) that involves chronic circadian disruption.

1. INTRODUCTION

Circadian disruption, a disturbance of biological timing, is increasingly common in modern society and has been associated with greater risk for obesity, diabetes, and other cardiometabolic disorders in cross-sectional studies (1–6). These disorders are also exacerbated by the consumption of unhealthy foods, frequent snacking, and eating in the late evening and night, which increase obesity and type 2 diabetes risk (7–9). There is accumulating evidence that these two behavioral risk factors may interact together to adversely affect metabolism (10–13): one study in mice reported that the addition of circadian misalignment to mice on a high-fat diet (HFD) caused greater glucose intolerance and insulin resistance than a HFD alone (12), and another recent study reported that circadian rhythm disruption in mice fed a HFD resulted in a diabetic state (14). However, no study has investigated how lower- and higher-fat diets interact differentially with chronic circadian disruption to impact glucose tolerance in both mice and humans.

While previous studies in humans have shown that sleep loss impairs glucose tolerance and insulin sensitivity (15–19), sleep restriction is often accompanied by circadian disruption through changes in the timing of light exposure, food intake, or waking behaviors (20). Moreover, circadian disruption itself may impair glucose regulation (21–29). In a paper from a related study (30), we reported no significant impairment of glucose tolerance from sleep restriction (~5 hours/night for 3 weeks) alone when circadian disruption was minimized, even when the sleep-deprived participants were fed a HFD. In this paper, we tested in both mice and humans the effect of HFD or lower-fat diet (LFD) during exposure to recurrent circadian disruption (RCD) that minimized sleep deficiency.

2. MATERIALS AND METHODS

2.1. Animal studies

2.1.1. Animals

We used young (2 months, n=32) and aged (22 months, n=38) male C57Bl6 mice from the National Institute of Aging mouse colony (Frederick, MD) for this study. Only male mice were included to avoid the potential confound of estrus cycling in female animals, which has previously been reported to impact glucose tolerance (31). Mice were procured at one or 20 months of age and were singly-housed in a temperature (mean±SD; 22±1°C)–and humidity (40%–60%)–controlled animal room maintained on a 12:12-h LD cycle. All mice had ad libitum access to standard chow diet (Research diets, 13% fat) and water during this initial segment.

2.1.2. Surgery and circadian and metabolic data collection

Mice were anesthetized with ketamine/xylazine (100 mg/kg and 10 mg/kg, IP) and implanted with radio-telemetry transmitters (TA-F10, Data Science International, St. Paul, MN) that allow simultaneous recording of Tb and LMA. Two weeks after surgery, mice were transferred and habituated for three days to the recording rooms and Tb and LMA data collection began. After five to seven days in 12:12 LD, RCD mice were then transferred to 10:10 LD cycles, while in parallel the Control mice were continued on the same 12:12 LD cycle (see Figure 1). Tb and LMA data were continuously collected every five minutes during the entire experimental duration (except on the days of glucose tolerance test [GTT] and insulin tolerance test [ITT]) using Dataquest ART 4.1 program (Data Sciences International, USA). Body weight and food intake in all mice were measured every two weeks. Glucose tolerance tests and insulin tolerance tests were performed on the days of maximal alignment of LD and LMA cycles during the ninth and 11th week of exposure. For GTT, mice were fasted overnight (16 h) and intraperitoneally (i.p) injected with 2 g/kg glucose (at ZT2), and blood glucose levels from tail blood samples were measured at 0, 15, 30, 60, and 120 minutes post-injection using a glucometer (One-touch, USA). Similarly, for the ITT, insulin (0.9 IU/kg, i.p, ZT2) was administered and glucose levels were measured at 0, 15, 30, 60, and 120 minutes post-injection. The 120-minute Area-Under-the-Curve (AUC) was calculated using the trapezoid method (GraphPad Software Inc., San Diego, CA).

Figure 1. Schematic illustration of animal study light-dark schedules.

Each horizontal bar represents 24 hours of the experiment, and subsequent days of the experiment are shown below the previous day. White and grey colors represent the light and dark episodes, respectively. While the study schedules were imposed for 12 weeks, only the first 6 days of the experiments are shown. (A) Light-dark schedule for the Control mice; (B) light-dark schedule for the RCD mice.

2.1.3. Diet

One set of Control and RCD mice in each age group received a LFD of regular chow (13% fat) while a second set received a HFD containing 45% fat (D12451, Research Diets, Inc, New Brunswick, NJ) throughout the entire 12-week experiment.

2.1.4. Circadian data analysis

Circadian rhythms in LMA and Tb were assessed by plotting using double raster plots and circadian period (τ) and amplitude were calculated using Clocklab software (Coulbourn Instruments, PA, USA). For LMA and Tb data, χ2 periodograms were performed using Clocklab for weeks six to eight at five-minute resolution. Finally, the total LMA counts and mean Tb per day for a 5-day period in controls and the equivalent 120-h period in RCD mice (starting from the day of maximal alignment) were calculated for comparing LMA and Tb changes across groups.

2.1.5. Sleep-wake recording and analysis

Under anesthesia (ketamine 100 mg/kg and Xylazine 10 mg/kg, IP), two additional groups of young mice were implanted with electroencephalogram (EEG) and electromyogram (EMG) electrodes (32) for EEG/EMG recordings. Two weeks after the surgery, mice were exposed to RCD or Control schedules for five weeks while on HFD. 24-hour EEG and EMG recordings from Control and RCD mice were performed on the sixth week of exposure. In RCD mice, these recordings were performed specifically on the day of maximal alignment (of LD and LMA cycle). EEG/EMG data for each mouse were divided into 12-second epochs and manually scored into one of three stages of sleep-wake: wake, non-rapid eye movement (NREM) sleep, or rapid eye movement (REM) sleep using previously described scoring criteria (33). Percentages of time spent in wake, NREM and REM sleep per 24 h, bout numbers and average bout durations of each of these stages were calculated.

2.2. Human studies

2.2.1. Participant recruitment and eligibility criteria

Healthy adult participants were recruited from the general community using online and newspaper advertisements, recruitment letters, and flyers. Participants completed a screening consisting of medical history evaluation, physical examination, electrocardiogram, clinical blood tests (complete blood count, comprehensive metabolic panel, thyroid function test), and urinalysis to rule out medical disorders; an all-night at-home polysomnogram to rule out sleep disorders; psychological questionnaires (MMPI, Beck Depression Inventory) and a semi-structured clinical psychological interview to rule out psychological disorders (34). Exclusion criteria included current or chronic medical conditions, regular use of prescription or over-the-counter medication, BMI>33 kg/m²; current or past psychiatric or psychological disorders; smoking; excessive caffeine consumption; regular night shift work within the last three years, travel across more than one time zone within three months, significant sleep complaints, and habitual sleep duration shorter than 6.6 hours or longer than nine hours. Participants were assigned to RCD or Control groups within each diet, stratified by gender and with a randomization block size of six. No participant was ever studied more than once under the same diet and experimental condition. Data collection procedures were standardized for all participants.

2.2.2. Study protocol and conditions

Participants maintained a self-selected consistent sleep-wake schedule with a 10-hour nighttime sleep opportunity for at least three weeks prior to admission. Compliance was verified by wrist actigraphy (Actiwatch-S, Philips-Respironics, PA, USA; or MotionWatch-8, CamNtech, Cambridge, UK), sleep diaries, and time-stamped voicemails at bedtime and wake time. Participants were instructed not to use drugs (prescription, over the counter, recreational), alcohol, nicotine, or caffeine during this pre-study segment; compliance was assessed with a urine toxicology screen upon study admission.

All participants were studied individually in a private room in the Intensive Physiological Monitoring Unit of the Center for Clinical Investigation at Brigham and Women’s Hospital. The laboratory environment was free of time cues and maintained at a temperature of 23.9+1.7°C (mean±SD), with complete darkness (<.02 lux) during scheduled sleep opportunities and standard room lighting (90 lux) during scheduled wake episodes. Participants were not permitted to exercise but could engage in sedentary activities such as reading, writing, watching movies, arts and crafts, or listening to music during free time. Research technicians observed participants by closed circuit television or direct observation during waking episodes to ensure that participants complied with the study protocol and remained awake. Study events were timed relative to each participant’s habitual rest-activity schedule, as documented in the week immediately prior to the inpatient portion of the protocol.

Upon admission, all participants were first scheduled to three sleep extension days, each of which included 16 hours of time in bed consisting of a 12-hour nighttime sleep opportunity and a four-hour nap opportunity scheduled in the middle of the wake episode. These sleep extension days were intended to minimize the effect of any prior history of sleep deficiency before the study intervention (26, 35). Sleep extension was followed by three Baseline (BL) days, each with a 10- (LFD) or 8- (HFD) hour nighttime sleep opportunity.

After the Baseline days, participants in the Control groups underwent three weeks with a 10- [Control (LFD)] or 8-hour [Control (HFD)] sleep opportunity each night. The 10-hour sleep opportunity was the control/baseline condition in previous CSR studies (26, 35–38). In the HFD group, the sleep opportunity for the three baseline nights was reduced to 8 hours to avoid metabolic effects from restricting the daily eating duration (39–42). Participants in both LFD and HFD RCD groups next underwent three weeks of exposure (EXP) to 28-hour “days” designed to induce recurrent circadian disruption, each with an 11.67-hour sleep opportunity (equivalent to 10 hours of sleep per 24 hours; the extended sleep opportunity was chosen to reduce or avoid inadvertent sleep restriction due to scheduling sleep at adverse circadian phases). Finally, all participants underwent nine days of recovery (REC), each with ten hours (LFD) or eight hours (HFD) of scheduled sleep opportunity to match baseline conditions (Fig. 4).

Figure 4. Human inpatient study schedules.

Study day is indicated along the left side and representative clock hour along the top axis. Solid black bars represent scheduled sleep episodes in darkness, while gray and white indicate parts of the study conducted under 4 lux and 90 lux lighting levels, respectively. Diagonal hatched bars indicate times during wake episodes when participants maintained a semi-recumbent posture in bed. The timing of standardized breakfast meal responses (green bars in Control group; blue bars in RCD group) for assessment of glucose and insulin is also indicated. Depicted schedules are for LFD groups.

2.2.3. Controlled diet

During the inpatient portion of the study, participants received either a lower- (LFD) or high-fat (HFD) controlled nutrient diet free of caffeine. LFD consisted of 58–60% carbohydrates, 15–17% protein, and 25–27% fat, with calories distributed evenly across breakfast, lunch, and dinner; during the three weeks of RCD, a snack consisting of 16.7% of the 24-hour calorie target was added after dinner. HFD consisted of 30–40% carbohydrates, 15–20% protein, and 45–50 % fat, with breakfast, lunch, and dinner each containing 25% of the daily calories, and the remaining 25% distributed evenly across two snacks served after lunch and dinner. A higher fat percentage than typically observed in the general population was chosen to maximize the effect of the exposure. Both diets included calcium 800–1350 mg, potassium 100 mEq (+/− 30%), and sodium 150 mEq (+/− 30%), and 2,000 mL fluid per 24 hours. The initial diet was calculated at the start of each study based upon the Mifflin St. Jeor equation (43) with an activity factor 1.3 (LFD) or 1.6 (HFD). Standardized breakfasts were identical within (but not between) each participant for all meal response testing, and participants were required to finish all food within 30 minutes.

2.2.4. Standardized breakfast meal response

Three meal response testing days were scheduled: during BL, at an aligned phase during the third week of EXP, and during REC. In the HFD groups only, two BL meal responses were conducted and averaged for the final analyses. In the Control (HFD) group, two REC meal responses were conducted and averaged together for the final analyses. On meal response days, two fasted blood samples were scheduled to be collected via an indwelling catheter in the morning 60 minutes and five minutes before the standardized breakfast was served. Blood samples were then scheduled to be taken every ten minutes for 60 minutes following the start of breakfast, then every 30 minutes during the following 120 minutes (hours 2–3 post-breakfast), and a final sample 240 minutes (4 hours) after breakfast. Samples were placed on ice, centrifuged within one hour of collection, aliquoted into separate tubes for glucose, insulin, and lipids, and either frozen at −80° C (glucose and insulin samples) or sent for assay within 48 hours of collection (lipid samples).

2.2.5. Glucose and insulin assays and analysis

Serum glucose was measured using a YSI 3300 STAT Plus Glucose and L-Lactate Analyzer (YSI Life Sciences) with a sensitivity of 3.6 mg/dL, an inter-assay precision CV of 3.3–6%, and an intra-assay precision CV of 1.4–1.8%. Serum insulin was assayed using a chemiluminescent immunoassay kit (Beckman Coulter, Inc., Fullerton, CA) with a sensitivity of 0.03 μIU/mL, an inter-assay precision of 3.1–6.6%, and an intra-assay precision of 3.0–4.3%. Fasting glucose and insulin values were calculated by averaging across samples scheduled to be drawn between 60 minutes and 5 minutes prior to serving the standardized breakfast. Glucose and insulin meal responses were quantified by calculating the Area Under the Curve (AUC) from postprandial minute 10 through postprandial minute 180; postprandial minute 240 was not included in the AUC calculation due to occasional missing values. Linear interpolation by method JOIN (SAS version 9.4, SAS Institute, Cary, NC) was used for other missing values.

2.2.6. Lipid assays and analysis

Fasted serum samples were collected ~60 minutes before the standardized breakfast on two to three BL days (study days 5–7), during EXP (44), and on two to three REC days (study days 34–36) and were sent to LabCorp to be assayed for triglycerides, total cholesterol, and high-density lipoprotein (HDL) cholesterol; low-density lipoprotein (LDL) cholesterol and very-low density lipoprotein (VLDL) cholesterol were estimated using the Friedewald equation (Lipid Panel, #303756, LabCorp, Burlington, NC). Values were calculated by averaging across samples within a study segment.

2.2.7. Body weight and body fat analysis

In the LFD groups, wake-time, fasted, post-void weight was measured using the same calibrated hospital scale and compared between BL, EXP, and REC days. In the HFD groups, weight was measured after breakfast to accommodate fasting energy expenditure measurements (45). Discovery W Dual-energy X-ray Absorptiometry (DXA) Scanner (Hologic, Bedford MA) was used to measure body fat composition prior to admission and after discharge. Scans prior to admission were performed within 15 days of admission and scans after discharge were performed within two days of discharge. The pre-study scan for one Control (LFD) participant was performed 2 months prior to their study due to delayed admission.

2.2.8. Polysomnographic recording and analysis

Ambulatory polysomnography (PSG) was recorded continuously throughout the study with a digital recorder (Vitaport3, Temec Technologies BV, Heerlen, Netherlands), except for a daily shower break. The montage included electroencephalography (EEG; C3, C4, Fz, Cz, Pz, Oz, referenced to linked mastoids M1 and M3), left and right electrooculography (EOG; LOC, ROC), bipolar submental electromyography (EMG; during sleep episodes only), and bipolar electrocardiograhy (ECGs). Filtered and digitized PSG data were visually scored by a registered PSG technologist in 30 second epochs in Vitascore (Temec Technologies BV) according to established criteria (46). A specialized in-house software program (TASCI File Manager) was used to organize the scored PSG data and to compile information from each sleep episode. Prorated average Total Sleep Time (TST) per 24 hours was calculated (e.g. TST per 28-hrs was scaled to the proportional amount of sleep per 24 hours) for the last two baseline sleep episodes, each of the RCD or control exposure sleep episodes, and each of the recovery sleep episodes. Scored PSG recordings for the four participants in the Control (LFD) group and for the REC segment in two participants of the HFD-RCD group were not available.

2.2.9. Circadian phase assessments

For each participant in the RCD groups, the intrinsic circadian period of the core body temperature data from the FD portion of the protocol was estimated using non-orthogonal spectral analysis (47). From this estimate of intrinsic circadian period, a circadian phase (from 0 to 359°) was assigned to each minute of the study, with 0° corresponding to the minimum of the waveform fit to the entire temperature data series. This information was used to select a meal response from the third week of exposure in the RCD groups that was aligned (i.e., at the same circadian phase as the baseline meal response (26).

2.3. Statistics

Statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC). For the animal data, metabolic outcomes were analyzed using a mixed model with three factors: age (young vs. older), diet (LFD vs. HFD), and condition (RCD vs. Control). All outcomes were approximately normally distributed. The critical significance level was set at α=0.05. The Bonferroni method was used to adjust for multiple comparisons [we considered 4 total comparisons: young RCD (HFD) vs. young Control (HFD), old RCD (HFD) vs. old Control (HFD), young RCD (LFD) vs. young Control (LFD), and old RCD (LFD) vs. old Control (LFD)], and both adjusted and unadjusted p values are reported. Data are reported in the text as least squares means unless otherwise specified. Figures depict raw means ± SEM.

Statistical analyses for the human data were performed using SAS version 9.4 (SAS Institute, Cary, NC). Although sample size and power calculations were originally computed based on two groups (Control vs. RCD), after a planned interim analysis of the data we changed the study protocol to include high- as well as low-fat diet, as informed by the animal results and in consultation with our external scientific advisory council recognizing that this would limit the number of participants in each of the four groups. Therefore, post-hoc analyses were used, showing that we have 69% power to detect an effect size of 1.46 computed based on the 180 min AUC glucose response (two-sided two-sample t-test, n=5, α =0.05). All metabolic outcomes were approximately normally distributed. Paired sample t-test or rank test was used to compare total body fat % prior to admission and after discharge and lipid levels between BL and REC within each group. Linear mixed models were used to analyze the outcomes measured repeatedly over time. Group [RCD (HFD), RCD (LFD), Control (HFD), Control (LFD)], condition (BL vs. EXP vs. REC), and group x condition were considered as fixed effects in the model, and subjects were considered as random effects. Gender, race (white vs. non-white), ethnicity (Hispanic vs. non-Hispanic), BMI, baseline weight, and AHI were tested as potential covariates. Only significant covariates were retained in the final model. We included all available data from one participant in the Control (HFD) group who was disempaneled on day 17. We were unable to obtain recovery meal response data for one RCD (HFD) participant due to IV problems. We do not have recovery sleep data for any Control (LFD) participants and two RCD (HFD) participants. Repeat participants were included as nested random effects. For most of the analyses, we only used observed data. Imputation methods such as linear interpolation and last observation carried forward were used when necessary for calculating AUCs, fasting glucose, and fasting lipids.

Data are reported in the text as least squares means (LSmeans), and a 95% bootstrap confidence interval, unless otherwise specified. Data presented in figures are raw means ± SEM. The critical significance level was set at α=0.05. The Bonferroni method was used to adjust for multiple comparisons (we considered 8 total comparisons: BL vs. EXP and BL vs. REC in each of the four groups) and both unadjusted and Bonferroni-adjusted p-values are reported.

2.4. Study approval

All animal experiments were conducted in accordance with the National Institutes of Health guidelines for the Care and Use of Laboratory Animals and were approved by the institutional animal care and use committee of Beth Israel Deaconess Medical Center (protocol #055-2016). All efforts were made to minimize the number of animals used and their suffering. All human experiments were reviewed and approved by the Partners Health Care Institutional Review Board (protocol 2014-P-000343). The study conduct adhered to the ethical principles outlined in the Declaration of Helsinki and each participant provided written informed consent. The human trial was registered on ClinicalTrials.gov (#NCT02171273).

3. RESULTS

3.1. Animal studies

3.1.1. Metabolic outcomes in young and old mice on low fat diet (LFD)

Young (n=8) and old (n=7) RCD (LFD) mice did not display any significant differences in weight gain (Figure 2A) or food intake (Figure 2B) when compared to young (n=8) and old (n=7) Control (LFD) mice, respectively. When measured on the day of maximal circadian alignment [of locomotor activity (LMA) and LD cycles], young but not old RCD (LFD) mice displayed higher fasting blood glucose levels compared to age-matched Control (LFD) mice (LSM: Fasting Glucose_RCD(LFD)=160.22 mg/dL; Fasting Glucose_Control(LFD)=131.75 mg/dL; LSM_diff = 28.47 mg/dL; unadjusted p=0.011, Bonferroni-adjusted p=0.045; Figure 2C). Neither young nor old RCD (LFD) mice showed significant decreases in glucose tolerance or insulin sensitivity compared to age-matched Control (LFD) mice. In response to a glucose load of 2 g/kg intraperitoneal (IP), Control (LFD) and RCD (LFD) mice exhibited similar increases in blood glucose levels and cleared glucose at a similar rate in both the young and old group (Figure 2D and 2E). Similarly, changes in blood glucose levels in response to 0.9 IU/kg insulin did not significantly differ between RCD (LFD) and Control (LFD) mice in either age group (Figure 2F and 2G).

Figure 2. Metabolic effects of RCD in young and old mice on LFD.

(A) Weight gain during the 12-week experimental interval, (B) daily food intake), and (C) fasting blood glucose levels in young (solid bars) and old (hashed bars) Controls on a 12:12-hour LD (green) and RCD mice on 10:10-hour LD cycle (blue). (D) Changes in blood glucose levels in response to 2 g/kg glucose IP (glucose tolerance) in Control and RCD mice, € the 120-minute blood glucose AUC. (F) Changes in blood glucose levels in response to insulin i.p (insulin sensitivity) in Control and RCD mice; (G) the 120-minute blood glucose AUC). P-values from mixed model ANOVAs are denoted with Bonferroni-adjusted p<0.05 *. Number of mice in each group: young RCD(LFD), n=8; old RCD(LFD), n=7; young Control(LFD), n=7; old Control(LFD), n=7.

3.1.2. Metabolic outcomes in young and old mice on high fat diet (HFD)

Young and old RCD(HFD) mice displayed an accelerated weight gain compared to young and old Control (HFD) mice (Figure 3A). During the 12-week experiment, young (n=8) RCD (HFD) mice gained significantly more weight compared to young (n=7) Control (HFD) mice (LSM: Weight Gain_RCD(HFD)=9.525 g; Weight Gain_Control(HFD)=5.757 g; LSM_diff =3.768 g; unadjusted p=0.004, Bonferroni-adjusted p=0.015). Similarly, old (n=7) RCD (HFD) mice gained significantly more than old (n=7) Control (HFD) mice (LSM: Weight Gain_RCD(HFD)=12.214 g; Weight Gain_Control(HFD)=7.029 g; LSM_diff =5.186 g; unadjusted p=0.0002, Bonferroni-adjusted p=0.0008). This weight gain occurred even though there was no significant difference in daily food intake between RCD (HFD) and Control (HFD) mice in either age group (Figure 3B).

Figure 3. Metabolic effects of RCD in young and old mice on HFD.

(A) Daily food intake, (B) weight gain during the three month-experiment, and (C) fasting blood glucose levels in young (solid bars) and old (hashed bars) Control mice on a 12:12 LD cycle (green) and RCD mice on 10:10 LD cycle (blue). (D) Changes in blood glucose levels in response to 2 g/kg glucose i.p (glucose tolerance) in Control and RCD mice, and (E) the 120-minute blood glucose AUC. (F) Changes in blood glucose levels in response to insulin i.p (insulin sensitivity) for Control and RCD mice; (G) the 120-minute blood glucose AUC. P-values from mixed model ANOVAs are denoted with Bonferroni-adjusted p<0.05 *. Number of mice in each group: young RCD(LFD), n=8; old RCD(LFD), n=7; young Control(LFD), n=7; old Control(LFD), n=7.

Mice in both age groups fed a HFD exhibited higher fasting blood glucose levels compared to age-matched controls (Figure 3C). Young RCD (HFD) mice had significantly higher fasting glucose than young Control (HFD) mice (LSM: Fasting Glucose_RCD(HFD)=163.5 mg/dL vs. Fasting Glucose_Control(HFD)=117 mg/dL; LSM_diff = 46.5 mg/dL; unadjusted p<0.0001, Bonferroni-adjusted p<0.0001), as did old RCD (HFD) mice compared with old Control (HFD) mice (LSM: Fasting Glucose_RCD(HFD)=183.71 mg/dL; Fasting Glucose_Control(HFD)=147 mg/dL; LSM_diff = 36.71 mg/dL; unadjusted p=0.0033, Bonferroni-adjusted p<0.013).

The change in glucose AUC in both young and old RCD (HFD) mice was not significant compared to Control (HFD) mice after Bonferroni adjustment. However, because in both age groups the RCD (HFD) mice exhibited a non-significant trend toward an increase in glucose AUC compared to Control (HFD) mice (Young: LSM: AUC_RCD(HFD)=46953 mg/dL·min vs. AUC_Control(HFD)=38880 mg/dL·min, LSM_diff = 8072 mg/dL·min; unadjusted p=0.044, Bonferroni-adjusted p = 0.176; Old: LSM: AUC_RCD(HFD)=47076 mg/dL·min vs. AUC_Control(HFD)=38032 mg/dL·min, LSM_diff = 9044 mg/dL·min; unadjusted p=0.034, Bonferroni-adjusted p = 0.136), we combined both age groups to increase our power to test for the difference between RCD and Control conditions under the HFD. With the combined age groups, we observed a significant increase in glucose AUC in response to the glucose tolerance test in RCD (HFD) animals compared to Control (HFD) animals (LSM: AUC_RCD(HFD)= 47014 mg/dL·min vs. AUC_Control(HFD)=38456 mg/dL·min, LSM_diff = 8558 mg/dL·min; unadjusted p=0.0041, Bonferroni-adjusted p = 0.0082).

Young RCD (HFD) mice did not show significantly different changes in blood glucose levels in response to IP injection of insulin in the insulin tolerance test (ITT) when compared to Control (HFD) mice (Figure 3F and 3G). In contrast, old RCD (HFD) mice exhibited a reduced response to insulin, and blood glucose levels remained significantly higher than Control (HFD) mice for the full 120 minutes after IP insulin injection, resulting in the AUC of the ITT curve being significantly greater in the RCD (HFD) than in Control (HFD) mice (LSM: AUC_RCD(HFD)=9133 mg/dL·min vs. AUC_Control(HFD)=6229 mg/dL·min, LSM_diff = 2904 mg/dL·min; unadjusted p=0.0035, Bonferroni-adjusted p = 0.014; Figure 3F and 3G).

3.1.3. Locomotor activity, temperature rhythms, and sleep in recurrent circadian disruption (RCD) mice

As shown previously (48), exposure to the 10:10 LD schedule caused body temperature (Tb) rhythms and locomotor activity (LMA) rhythms to continue to oscillate with a period of >24 h (Figure S1 and Table S1). A few mice exhibited a period of <24 h during the first three to four weeks, then changed into stable >24 h Tb and LMA rhythms.

As both young and old RCD (HFD) mice showed dramatic weight gain and deficits in glucose homeostasis, and previous studies showed a causal link between chronic sleep loss and metabolic syndrome in humans, we investigated whether chronic sleep loss in the older RCD (HFD) mice could have contributed to the observed adverse metabolic effects. We collected sleep-wake data from the mice on the day of maximal alignment during the sixth week of exposure. Percentages of wake, non-Rapid Eye Movement (NREM) and Rapid Eye Movement (REM) sleep as well as total bout numbers and average duration of these states in RCD (HFD) mice (n=6) were not significantly different from Control (HFD) mice (n=5, Figure S1), suggesting that RCD (HFD) mice experience neither chronic sleep loss nor sleep fragmentation (i.e., sleep episodes with multiple awakenings) on this protocol, consistent with previous findings (49). Collectively, the data from our mouse model of RCD indicate that RCD even without sleep loss may accelerate weight gain and produce deficits in glucose homeostasis irrespective of age, but only when the mice were fed a HFD.

3.2. Human Studies

3.2.1. Human study schedules and participant characteristics

We performed a 37-day inpatient trial of healthy adult volunteers consisting of three 24-hour days of sleep extension [16 hours of time-in-bed (TIB)], three 24-hour baseline days (BL, 10 or 8 hours TIB in LFD and HFD groups, respectively), three weeks of RCD exposure (EXP, 11.67 hours TIB per 28-hour cycle, equivalent to 10 hours/24 hours), and nine days of recovery (REC, 10 or 8 hours TIB on a 24-hour day in LFD and HFD groups, respectively). We also studied participants on a Control schedule under both LFD and HFD who spent the same amount of time in the laboratory with the same sleep extension, BL, and REC segments, but who remained on the BL sleep-wake schedule for the remaining four weeks of the study (corresponding to the EXP and REC segments; Figure 4).

A total of 19 studies were conducted in 13 participants. One RCD (LFD) participant (66-year-old male) quit the study on day 9, and one RCD (HFD) (58 years, female) participant was disempaneled on day 24 after developing a UTI; data from these two participants were not included in analyses. Thus, data reported are derived from 17 studies conducted in 12 participants: five studies in the RCD HFD) group (n=5; 55.8±10.5 years, 38–65 years, 2 women), four studies in the RCD (LFD) group (n=4; mean±SD; 61.3±6.0 years, range 56–69 years, 3 women), four studies in the Control (LFD) group (n=4; 57.0±2.4 years, 55–60 years, 2 women), and four studies in the Control (HFD) group (n=4; 52.0±12.1 years, 34–60 years, 1 woman). All female participants were post-menopausal. See Table S2 for additional participant characteristics. Two participants completed the trial under both RCD (LFD) and RCD (HFD) conditions, one participant completed the trial under both Control (LFD) and Control (HFD) conditions, one participant completed the trial under both RCD (HFD) and Control (HFD) conditions, and one participant completed the trial three times under RCD (LFD), RCD (HFD), and Control (LFD) conditions. No participant completed the trial more than once under the same diet/experimental group. No repeat participant was studied within 6 months of their previous study.

3.2.2. Metabolic outcomes in human participants on lower fat diet (LFD)

In the RCD (LFD) group, there were no significant changes in postprandial glucose 180-minute AUC values following a standardized breakfast meal when comparing the BL, EXP, and REC time points (Figure 5A and 5E). Similarly, the Control (LFD) group showed no significant changes in glucose AUC values when comparing the BL, EXP, and REC time points (Figure 5B and 5F).

Figure 5. Postprandial glucose and insulin in humans on LFD.

Postprandial glucose and insulin profiles are depicted following a standardized breakfast meal at baseline (BL) and after exposure (EXP) to the RCD (panels A, C) or Control schedule (panels B, D). AUC values over the first 180 minutes following the meal at BL, EXP, and REC are shown for the RCD group (panels E, G) and Control group (panels F, H). Values shown are mean±SEM. Bonferroni-adjusted p-values from LSmeans post-hoc test are depicted as follows: p≤0.05 *, p≤0.01 **, p≤0.001 ***. Number of participants in each group: RCD(LFD), n=4; Control(LFD), n=4.

In the RCD (LFD) group, there was a significant reduction in postprandial insulin 180-minute AUC values in response to a standardized breakfast meal during EXP when compared to BL (Figure 5C and 5G; LSM: AUC_BL=9972 μIu/mL·min vs. AUC_EXP=6931 μIu/mL·min; LSM_diff: 3041 with a 95% bootstrapping CI of (875, 1700), unadjusted empirical bootstrapping p=0.007, Bonferroni-adjusted p=0.056). REC value for postprandial insulin AUC was also lower than the BL value (LSM: AUC_BL=9972 μIu/mL·min vs. AUC_REC=7322 μIu/mL·min; LSM_diff: 2650 with a 95% bootstrapping CI of (909, 967), unadjusted empirical bootstrapping p=0.006, Bonferroni-adjusted p=0.048). In the Control (LFD) group there were no significant changes in insulin AUC values when comparing EXP to BL or REC to BL (Figure 5D and Figure 5H).

Finally, we saw no significant changes in fasting glucose levels in either the RCD (LFD) or Control (LFD) groups when comparing the BL, EXP and REC time points (Table S3).

3.2.3. Metabolic outcomes in human participants on high fat diet (HFD)

In the RCD (HFD) group, there were significant changes in postprandial glucose 180-minute AUC values in response to a standardized breakfast meal during EXP when compared to BL (Figure 6A and 6E). Post-hoc comparisons revealed that after two to three weeks of RCD, the EXP postprandial glucose AUC value was significantly higher than the BL AUC value (LSM: AUC_BL=17049 mg/dL·min vs. AUC_EXP=18553 mg/dL·min; LSM_diff: −1504 with a 95% bootstrapping CI of (−1909, −1063), unadjusted empirical bootstrap p<0.0001, Bonferroni-adjusted p<0.0001). REC postprandial glucose AUC value was not significantly different than BL AUC value. In the Control (HFD) group, there were no significant changes in postprandial glucose AUC values when comparing EXP to BL or REC to BL (Figure 6B and 6F).

Figure 6. Postprandial glucose and insulin in humans on HFD.

Postprandial glucose and insulin profiles are depicted following a standardized breakfast meal at baseline (BL) and after exposure (EXP) to the RCD (panels A, C) or Control schedule (panels B, D). AUC values over the first 180 minutes following the meal at BL, EXP, and REC are shown for the RCD group (panels E, G) and Control group (panels F, H). Values shown are mean±SEM Bonferroni-adjusted p-values from LSmeans post-hoc test are depicted as follows: p≤0.05 *, p≤0.01 **, p≤0.001 ***. Number of participants in each group: RCD(HFD), n=5; Control(HFD), n=4.

We found no significant changes in postprandial insulin AUC values in the RCD (HFD) group (Figure 6C and 6G) or in the Control (HFD) group (Figure 6D and 6H) when comparing the BL, EXP, and REC time points. Finally, we saw no significant changes in fasting glucose levels in either the RCD (HFD) or Control (HFD) groups when comparing the BL, EXP, and REC time points (Table S3). For all individual meal response time courses, see Figure S2 and Figure S3.

3.2.4. Body weight, body fat, fasting lipids, and sleep in human participants

There was a significant change in body weight in the RCD (LFD) and the Control (LFD) groups when comparing the BL, EXP, and REC time points. Post-hoc comparison revealed that participants in the RCD (LFD) group lost 1.1±0.2 kg (~1.4%) of body weight from BL to REC (LSM_diff: 1.1 with a 95% bootstrapping CI of (0.6,1.5), unadjusted empirical bootstrap p=0.002, Bonferroni-adjusted p=0.016), and significant change in body weight from BL to EXP (LSM_diff: 0.4 with a 95% bootstrapping CI of (0.2,0.6), unadjusted empirical bootstrap p=0.001, Bonferroni-adjusted p=0.008). Participants in the Control (LFD) group lost weight from BL to EXP, but the difference was not significant after adjustment (LSM_diff: 1.0 with a 95% bootstrapping CI of (0.5, 1.8), unadjusted empirical bootstrap p=0.027, Bonferroni-adjusted p=0.218). They also lost 1.1±0.3 kg (~1.5%) of body weight from BL to REC (LSM_diff: 1.1 with a 95% bootstrapping CI of (0.6, 1,5), unadjusted empirical bootstrap p=0.001, Bonferroni-adjusted p=0.008). Participants in the RCD (HFD) group showed an increase in body weight from BL to REC but the difference was not significant after adjustment (LSM_diff: −0.4 with a 95% bootstrapping CI of (−0.6, −0.1), unadjusted empirical bootstrap p=0.024, Bonferroni-adjusted p=0.196). There was no significant change in body weight in the Control (HFD) group.

Dual-energy X-ray absorptiometry scans performed prior to admission and after discharge revealed no significant changes in percentage of total body fat in the RCD (LFD), Control (LFD), or Control (HFD) groups (Table S3). In the RCD (HFD) group, the only group to show a trend towards an increase in weight, there was a significant increase in total body fat (percent of body mass) from pre-study to post-study (Table S3, LSM_diff: 1.0 with a 95% bootstrapping CI of (0.6–1.3), empirical bootstrap p=0.022).

There were no significant changes in fasting lipid levels when comparing BL and REC time points in any group (Table S3).

To verify that sleep loss was minimized on the RCD protocol, we calculated 24-hour prorated averages (see Methods) for total sleep time during BL, EXP, and REC from polysomnography (PSG) recordings in the RCD (LFD) and RCD (HFD) groups and found no significant differences in average total sleep time when comparing the BL, EXP, and REC segments (Table S3). We also found no significant change in sleep between BL and EXP in the Control (HFD) and Control (LFD) groups.

4. DISCUSSION

We found that RCD led to increased weight gain and impaired glucose tolerance and insulin sensitivity in young and old mice, but only when combined with HFD. Similarly, humans also showed impaired glucose tolerance following a history of RCD while minimizing sleep loss, but only when combined with HFD. We did not detect impaired glucose tolerance after chronic exposure to RCD in either mice (12 weeks) or humans (two to three weeks) on LFD, suggesting that RCD alone was not sufficient to affect glucose homeostasis.

In mice fed an HFD, RCD (compared to Control conditions) increased fasting glucose levels and impaired responses to glucose and insulin challenges, even when these were assessed at a re-aligned circadian phase. In contrast, we did not observe an effect of RCD on responses to glucose or insulin challenges in mice fed a LFD. We observed this overall trend in both young and older mice, although older mice undergoing RCD with HFD exhibited greater variance in response to the glucose tolerance test. Interestingly, older but not younger mice fed an HFD also exhibited a reduced response to the insulin tolerance test compared to controls, suggesting that age may influence the effect of RCD on insulin sensitivity. Further studies are needed to explore this potential effect of age.

While a previous study in mice found glucose intolerance and insulin insensitivity in mice exposed to both HFD and circadian disruption (12), it did not include a comparison of mice on LFD either with or without circadian disruption. Thus, the importance of the HFD for inducing metabolic consequences during circadian disruption was not demonstrated in that study. Similarly, we observed in humans an increase in glucose levels in response to a standardized breakfast when assessed at an aligned circadian phase, following exposure to RCD in participants on an HFD but not on a LFD. Moreover, in contrast to some previous studies of high-fat diet that have found adverse metabolic effects of high-fat diet in certain populations (50, 51), we did not observe changes in glucose tolerance in Control participants on an HFD, suggesting that three to four weeks of HFD consumption without circadian disruption may not by itself impair glucose tolerance when body weight is maintained in healthy participants. Although human participants on the LFD had higher absolute peak glucose levels than those on the HFD, these differences between the groups may have resulted from the lower carbohydrate content of the HFD standardized meal (52). The change in glucose tolerance from baseline after exposure was significant only in RCD participants on an HFD.

The metabolic effects seen in mice exposed to HFD and RCD were accompanied by weight gain that exceeded that seen in Control mice on the HFD, as has been previously reported (53, 54). There were no differences in caloric intake, overall locomotor activity, or body temperature between the mouse groups, suggesting that changes in energy expenditure (26) or gut microbiome (55) may underlie this effect. However, young mice exposed to RCD and LFD also exhibited increased fasting glucose levels despite showing no change in weight, suggesting an adverse effect of RCD on glucose metabolism independent of weight gain. Furthermore, we observed impaired glucose tolerance without weight gain in human participants exposed to RCD and HFD, despite controlling for environmental and dietary effects on weight. Previous studies of time-restricted feeding have reported improvements in insulin and/or glucose control when feeding takes place earlier in the day (i.e. better aligned with circadian rhythms) or during the active phase, and these improvements have been reported even in the absence of weight loss, suggesting that the circadian system may influence glucose homeostasis without changes in body weight (39, 56).

Interestingly, evidence from recent animal studies suggests that HFD may itself induce persistent disruption of the adipose circadian rhythm, and that this circadian disruption may play a key role in the development of subsequent obesity (57–59). In our study, only human participants in the RCD (HFD) group showed a significant increase in body fat (as measured by DXA scans) and a trend toward increased body weight.

Surprisingly, we did not find in either mice or humans any significant impairment of glycemic control during RCD while minimizing sleep loss on a LFD, consisting of 13% fat (regular chow) in mice or 25–27% fat in humans—comparable to the proportion of fat in many low-fat dietary interventions for cardiometabolic disorders (60). We previously found that acute and short-term recurrent circadian misalignment impaired both glucose tolerance and insulin sensitivity in humans on a similar LFD (21–25), contrary to the absence of adverse metabolic effects in the RCD(LFD) group in the current study. In fact, in the current study we observed lower insulin levels in response to a standardized meal in human participants after exposure to RCD on a LFD. We previously observed a similar reduction in insulin secretion in participants on a LFD who had RCD combined with sleep restriction for 2–3 weeks (26), an outcome which we attributed to inadequate insulin secretion. However, the RCD (LFD) participants in the current study did not exhibit increases in glucose despite the reduction in insulin secretion, raising an alternative possibility that this reflects some degree of internal desynchrony resulting from the history of circadian disruption, which leads to the uncoupling of the insulin response from normal glucose control (24), a speculative hypothesis that would require further study.

Another possible reason for this discrepancy is that we minimized sleep loss during RCD in the present study by extending the duration of time in bed. Other in studies humans in which circadian disruption is combined with sleep restriction have found impairments in both glucose tolerance and insulin sensitivity when participants consume a similar LFD (26, 61). However, as we have shown elsewhere(30), chronic sleep restriction without concurrent circadian disruption also does not account for the observed metabolic effects. Taken together, these results suggest that the effect of a LFD on glucose metabolism in response to RCD is insufficient to overcome the adverse effect of RCD combined with chronic sleep restriction. Interestingly, we found significantly lower levels of insulin following a test meal in participants consuming the LFD during chronic sleep restriction when circadian disruption was minimized (30). These results are consistent with animal studies, which have found that sleep loss causes weight loss, not weight gain, and does not affect glucose tolerance (33, 62–65), and with some human studies that have found that sleep restriction increases energy expenditure (66). Thus, our current results suggest that RCD increases susceptibility to obesity in mice and impaired glucose tolerance in both mice and humans and leads to adverse metabolic consequences when combined with additional insults such as a HFD or chronic sleep loss.

Limitations:

Our sample of healthy adult humans was small and consisted mainly of men and women over the age of 50. All of the women were post-menopausal, so menstrual variation was not a factor. However, generalizing from this small and relatively homogeneous group to a more diverse population will require further study. Some participants were studied multiple times, but we were careful to ensure that no participant ever repeated exactly the same experimental and diet conditions. We believe that these repeat participants do not confound our results, but rather strengthen our findings, as several of the same individuals were studied under different conditions; therefore, any different responses observed would be the result of condition rather than inter-individual differences. However, repeatedly studying the same individual is not ideal (unless done consistently, for example in the context of a crossover study) as it inevitably impacts the statistical variance of the outcomes.

The sample of mice was based on a power analysis for expected differences and included 8–9 per group. However, the modest unadjusted p-values, some of which became non-significant with Bonferroni correction, suggests that it would be useful in the future to use a larger sample of mice as well. In addition, we used a single inbred strain of mice (C57BL6/J) which is very commonly used for laboratory experiments involving mouse genetics, but which is known to have a deficit in melatonin secretion. Applicability to an outbred population would be important to study in the future. Further, to avoid additional variance due to estrus cycles, we studied only male animals in these experiments. Thus, it will be important to study female mice as well in the future.

Given the modest weight loss in the human RCD (LFD) group, it is possible that the LFD was hypocaloric compared to the participant’s home diet. However, this modest change did not confound the comparison with glucose regulation in the Control (LFD) group, which experienced a comparable weight loss.

A possible confound was that the duration of time over which human participants ate each day was shorter on the RCD (LFD) (12.33 h between first and last meal each day) than it was on the RCD (HFD) (15 h), which could have contributed to the effect of the LFD on glucose regulation in the RCD (LFD) group (39, 67–69). However, glucose regulation following exposure to RCD was compared in each participant to their baseline responses, so these between-group differences in the daily duration of meal consumption cannot account for the lack of a response to RCD observed among participants in the LFD condition. Moreover, our findings in humans were consistent with those in the mice, where there were no changes in feeding duration. In addition, the daily eating duration in the human Control groups on both diets were matched to the RCD groups; therefore, neither the diet nor the eating duration alone can account for the adverse metabolic impact of RCD combined with HFD.

Finally, while we extended the sleep opportunity during both the RCD (LFD) and the RCD (HFD) conditions to minimize sleep loss in the participants, some sleep fragmentation is likely to have occurred. However, the average sleep duration was comparable between both groups and cannot account for adverse effect on glucose metabolism observed in the RCD (HFD) group as compared to the RCD (LFD) group.

5. Conclusions

We found that only when combined with a HFD did RCD lead to weight gain and impaired insulin secretion in mice, and impaired glucose tolerance in both mice and humans. We found no adverse glycemic effects of RCD in mice or humans on a LFD. These are some of the longest, carefully controlled laboratory studies conducted in parallel in both mice and humans investigating the effects of a history of chronic circadian disruption on glucose metabolism. Our “exposure” assessments were performed at a re-aligned circadian phase by scheduling them at a similar time relative to the circadian rhythm of core body temperature. Thus, our findings demonstrate that prior exposure to circadian disruption continues to disrupt glucose regulation even when realigned with respect to a central phase marker. While we were not able to assess peripheral rhythms in this study, this finding suggests that peripheral metabolic rhythms may require longer to resynchronize and/or may not remain synchronized to the centrally-driven rhythm of core temperature. Together with our other recent finding indicating that sleep restriction when minimizing circadian disruption does not have a significant adverse impact on glucose tolerance (30), our results suggest that the current widespread circadian disruption induced by irregular schedules, exposure to light at night, and access to unlimited and inexpensive high-fat foods may be a critical combination that contributes to the high prevalence of obesity and diabetes in industrialized societies. Future studies of the interaction of circadian disruption (of centrally-controlled rhythms as well as peripheral rhythms) and dietary factors are warranted for development of diet-based interventions for shift workers and those with irregular schedules.

HIGHLIGHTS.

• Recurrent circadian disruption combined with a high-fat diet caused significant weight gain and significantly impaired glucose tolerance and insulin sensitivity in young and older mice

• Recurrent circadian disruption combined with a high-fat diet significantly impaired glucose tolerance in humans

• Recurrent circadian disruption combined with a lower fat diet had no adverse effect on glucose tolerance or insulin sensitivity in either mice or humans

ABBREVIATIONS:

AUC

area under the curve

BL

baseline

CSR

chronic sleep restriction

EXP

exposure

FD

forced desynchrony

GTT

glucose tolerance test

HFD

high-fat diet

ITT

insulin tolerance test

LD

light-dark

LFD

lower-fat diet

LMA

locomotor activity

RCD

recurrent circadian disruption

REC

recovery

Tb

body temperature

ZT

zeitgeber time

Data and materials availability:

Execution of a materials transfer agreement is required by our institution for transfer of data.