Endogenous circadian system and circadian misalignment impact glucose tolerance via separate mechanisms in humans
- aMedical Chronobiology Program, Division of Sleep and Circadian Disorders, Brigham and Women’s Hospital, Boston, MA 02115;
- bDivision of Sleep Medicine, Harvard Medical School, Boston, MA 02115;
- cDepartment of Biobehavioral Health, Pennsylvania State University, University Park, PA 16802; and
- dOregon Institute of Occupational Health Sciences, Oregon Health & Science University, Portland, OR 97239
See allHide authors and affiliations
Edited by Joseph S. Takahashi, Howard Hughes Medical Institute, University of Texas Southwestern Medical Center, Dallas, TX, and approved March 20, 2015 (received for review October 1, 2014)

Significance
It is established that glucose tolerance decreases from the morning to the evening, and that shift work is a risk factor for diabetes. However, the relative importance of the endogenous circadian system, the behavioral cycle (including the sleep/wake and fasting/feeding cycles), and circadian misalignment on glucose tolerance is unclear. We show that the magnitude of the effect of the endogenous circadian system on glucose tolerance and on pancreatic β-cell function was much larger than that of the behavioral cycle in causing the decrease in glucose tolerance from morning to evening. Also, independent from circadian phase and the behavioral cycle, circadian misalignment resulting from simulated night work lowered glucose tolerance—without diminishing effects upon repeated exposure—with direct relevance for shift workers.
Abstract
Glucose tolerance is lower in the evening and at night than in the morning. However, the relative contribution of the circadian system vs. the behavioral cycle (including the sleep/wake and fasting/feeding cycles) is unclear. Furthermore, although shift work is a diabetes risk factor, the separate impact on glucose tolerance of the behavioral cycle, circadian phase, and circadian disruption (i.e., misalignment between the central circadian pacemaker and the behavioral cycle) has not been systematically studied. Here we show—by using two 8-d laboratory protocols—in healthy adults that the circadian system and circadian misalignment have distinct influences on glucose tolerance, both separate from the behavioral cycle. First, postprandial glucose was 17% higher (i.e., lower glucose tolerance) in the biological evening (8:00 PM) than morning (8:00 AM; i.e., a circadian phase effect), independent of the behavioral cycle effect. Second, circadian misalignment itself (12-h behavioral cycle inversion) increased postprandial glucose by 6%. Third, these variations in glucose tolerance appeared to be explained, at least in part, by different mechanisms: during the biological evening by decreased pancreatic β-cell function (27% lower early-phase insulin) and during circadian misalignment presumably by decreased insulin sensitivity (elevated postprandial glucose despite 14% higher late-phase insulin) without change in early-phase insulin. We explored possible contributing factors, including changes in polysomnographic sleep and 24-h hormonal profiles. We demonstrate that the circadian system importantly contributes to the reduced glucose tolerance observed in the evening compared with the morning. Separately, circadian misalignment reduces glucose tolerance, providing a mechanism to help explain the increased diabetes risk in shift workers.
In healthy humans, there is a strong time-of-day variation in glucose tolerance, with a peak in the morning and a trough in the evening and night (1⇓⇓⇓⇓–6). Understanding the underlying mechanisms of the day/night variation in glucose tolerance is important for diurnally active individuals as well as shift workers, who are at increased risk for developing type 2 diabetes (7⇓–9). The endogenous circadian system and circadian misalignment (i.e., misalignment between the endogenous circadian system and 24-h environmental/behavioral cycles) have been shown to affect glucose metabolism (4, 10⇓⇓⇓–14). However, the relative and separate importance of the endogenous circadian system and circadian misalignment—after accounting for behavioral cycle effects (including the sleep/wake, fasting/feeding, and physical inactivity/activity cycles, etc.)—on 24-h variation in glucose tolerance is not well understood.
Most species have evolved an endogenous circadian timing system that optimally times physiological variations and behaviors relative to the 24-h environmental cycle (15⇓–17). The mammalian circadian system is composed of a central pacemaker in the suprachiasmatic nucleus (SCN) of the hypothalamus along with circadian oscillators in virtually all tissues and organs of the body, also referred to as peripheral oscillators (15⇓–17). The molecular genesis of circadian rhythms by the SCN and peripheral oscillators involves transcription-translation feedback loops (15⇓–17). This system is entrained to the solar day by external photic (i.e., the light/dark cycle) and nonphotic (e.g., nutrient intake) inputs (15⇓–17). Causal evidence for an independent and important impact of the endogenous circadian system on glucose metabolism comes from animal and human studies. Rodents with SCN lesion, whole-body clock gene disruption, or organ-specific (e.g., pancreas) clock gene disruption have marked metabolic phenotypes, including impaired glucose tolerance, decreased β-cell function, reduced insulin sensitivity, hyperglycemia, and/or hyperinsulinemia (18⇓⇓⇓⇓–23). Moreover, rats display a 24-h glucose rhythm even when feeding is uniformly distributed across the light/dark cycle, showing the rhythm is independent of the fasting/feeding cycle, and it is abolished by SCN lesion (24). An endogenous circadian rhythm in circulating glucose and insulin levels has been detected in humans by using intensive protocols performed under constant conditions (“constant routine” protocols) or when the behavioral cycle influences are accounted for by evenly distributing them across the entire circadian cycle (“forced desynchrony” protocols) (4, 10⇓–12, 14). In addition, circadian misalignment—as occurs in shift workers—also leads to impaired glucose metabolism in rodents and humans (10, 25⇓⇓⇓⇓⇓⇓–32).
We are aware of no study to date that has systematically tested the relative and separate effects of circadian phase [biological morning (defined here as the endogenous circadian phase equivalent to ∼8:00 AM) vs. biological evening (∼8:00 PM)] and of circadian misalignment, independent of the behavioral cycle, on glucose metabolism in humans. To address these separate effects (Fig. 1), we assessed—by using a within-participant, cross-over design—glucose tolerance in response to identical mixed meals given at 8:00 AM and 8:00 PM when the behavioral cycle of participants was aligned or misaligned with their endogenous circadian system using a rapid 12-h shift of the behavioral cycle (Fig. 2). Together, these two protocols (aligned vs. misaligned) allow the separate assessment of behavioral and circadian influences by evenly scheduling behavioral factors (e.g., sleep/wake and fasting/feeding) relative to two distinct circadian phases. In addition, the protocols allow the separate assessment of the impact of circadian misalignment by comparing responses to test meals when the behavioral cycle is aligned vs. misaligned with the phase of the circadian system when such meals are normally consumed. We also tested whether the separate effects of the circadian system and circadian misalignment on glucose tolerance would subside or become amplified upon repeated daily exposure to circadian misalignment—as is typical for many shift workers—by determining any change from test day 1 to test day 3 (Fig. 2).
Schematic diagram of the separate effects of the endogenous circadian system, the behavioral cycle, and circadian misalignment (interaction between the circadian cycle and behavioral cycle) on glucose tolerance. In addition, our analysis tested whether the effects of the endogenous circadian system, behavioral cycle, and circadian misalignment on glucose tolerance were dependent on circadian misalignment exposure duration (acute vs. repeated).
Circadian alignment protocol (Top) and circadian misalignment protocol (Bottom). On day 1 in both protocols, participants received an ad libitum lunch at ∼12:00 PM. Caloric intake was prorated for the 12-h behavioral cycle on day 4 of the circadian misalignment protocol (i.e., they received 50% of the caloric content compared with the 24-h days). Light level was also 90 lux during test meal assessments. The letters B and D indicate breakfast and dinner, respectively. Numbers following B or D indicate test days (first or third), and letters following these numbers indicate whether the test meals were consumed during the circadian alignment (A) or circadian misalignment (M) protocol. To graphically represent the independent effects of the behavioral cycle, circadian phase and circadian misalignment in the subsequent figures, we (i) averaged breakfast time (BA and BM) and dinner time (DA and DM) test meal values separately across both protocols for each test day (behavioral cycle effect); (ii) averaged 8:00 AM (BA and DM) and 8:00 PM (DA and BM) test meal values separately across both protocols for each test day (circadian phase effect); and (iii) averaged alignment (BA and DA) and misalignment (BM and DM) test meal values within each protocol for each test day (circadian misalignment effect).
Results
We first report the main effects of the behavioral cycle (breakfast vs. dinner), circadian phase (biological morning vs. biological evening), or alignment condition (circadian alignment vs. circadian misalignment) on our outcomes. Following this, we report whether these effects are dependent on circadian misalignment exposure duration (test day 1 vs. test day 3).
Effect of Circadian Misalignment on 24-h Melatonin and Cortisol Profiles.
The results for 24-h melatonin and cortisol profiles are shown in SI Appendix, Fig. S1. The 24-h average melatonin level was 56% lower in the circadian misalignment protocol than in the alignment protocol (P < 0.0001). This effect was not statistically dependent on circadian misalignment exposure duration (P = 0.050); melatonin was reduced by 52% on test day 1 (P < 0.0001) and by 59% on test day 3 (P < 0.0001). Circadian misalignment had no effect on average 24-h cortisol level regardless of misalignment exposure duration (both P ≥ 0.39). As expected, circadian misalignment changed the timing of the melatonin and cortisol rhythms relative to the behavioral cycle (both P < 0.0001). Specifically, while misaligned, melatonin levels peaked around the middle of the wake period (rather than during the sleep opportunity in the aligned condition) and cortisol levels peaked around the end of the wake period (rather than at the beginning of the wake period in the aligned condition). These effects were dependent on the duration of exposure to circadian misalignment (both P ≤ 0.007), with the melatonin and cortisol rhythms being more blunted as well as delayed relative to the behavioral cycle in the third vs. the first test days, presumably because of a slight phase shift in the central circadian pacemaker. Because melatonin, unlike cortisol, is little affected by behaviors, the phase of the central circadian pacemaker was estimated by the peak time of circulating melatonin. Despite the clear acute suppressive effect of melatonin by light during the scheduled wake times under misaligned conditions, the timing of the peak could still be accurately assessed. The circular mean (circular variance) peak clock times (determined by nonorthogonal spectral analysis) for melatonin in the alignment condition were 3:36 AM (0.01) on test day 1 and 3:58 AM (0.02) on test day 3, whereas, for the misalignment condition, mean peak clock times were 4:55 AM (0.16) on test day 1 and 7:46 AM (0.36) on test day 3. Based on this estimate, the circadian phase at which the 8:00 AM and 8:00 PM test meals were given on test day 1 were very similar between the aligned and misaligned protocols (differing by an average of only 1 h 18 min). By test day 3, the estimated phase difference had increased by 2 h 29 min (to 3 h 47 min). For this reason, we focused on test day 1 when reporting on the effect of circadian phase (in case of significant misalignment exposure duration effect).
Behavioral Cycle Effects, Independent of Circadian Effects: Glucose Tolerance and Early-Phase Insulin Were Lower at Dinner Time than at Breakfast Time.
The results for the impact of the behavioral cycle on glucose tolerance and insulin responses to tests meals are shown in Figs. 3 and 4 and SI Appendix, Figs. S2–S5. Two-hour postprandial glucose area under the curve (AUC) was 8% higher at dinner time than breakfast time (P < 0.0001), reflecting relatively reduced glucose tolerance at dinner. Similar results were found for peak postprandial glucose. Early-phase postprandial insulin AUC was 14% lower at dinner time than breakfast time (P = 0.003), suggesting reduced β-cell function at dinner time. Late-phase postprandial insulin AUC was 14% higher at dinner time than breakfast time (P < 0.0001), suggesting reduced insulin sensitivity at dinner time (as evidenced by dinner time postprandial glucose being higher despite late-phase postprandial insulin also being higher). Similar findings were found for early- and late-phase postprandial insulin secretion rate (ISR). Exposure duration to misalignment did not change the impact of the behavioral cycle on postprandial glucose, insulin, or ISR (all P ≥ 0.066).
Effects of the behavioral cycle (Left), circadian phase (Middle), and circadian misalignment (Right) on postprandial glucose and insulin profiles. Data are derived from eight identical test meals given at 8:00 AM or 8:00 PM in the circadian alignment and misalignment protocols. Data are derived as described in the legend of Fig. 2. Black bars represent 20-min test meals. Statistical comparisons between these conditions are presented in Fig. 4. Data are presented as mean ± SEM.
Effects of the behavioral cycle (Left), circadian phase (Middle), and circadian misalignment (Right) on postprandial glucose and early- and late-phase insulin AUCs. Data are derived as described in the legend of Fig. 2. Probability values: behavioral cycle, breakfast vs. dinner; circadian phase, biological morning vs. biological evening; alignment condition, circadian alignment vs. circadian misalignment; interaction with test day indicates if the abovementioned comparisons were dependent on circadian misalignment exposure duration (test day 1 vs. test day 3). Data are presented as mean ± SEM.
No Effect of the Behavioral Cycle on Fasting Glucose or Insulin.
The behavioral cycle results for fasting glucose, insulin, and ISR are shown in SI Appendix, Figs. S3 and S6. Fasting glucose and insulin were not different between breakfast and dinner (both P ≥ 0.40). However, ISR was 3% higher at dinner time than breakfast time (P = 0.019). Exposure duration to circadian misalignment did not change these effects (all P ≥ 0.41).
Circadian Effects, Independent of Behavioral Effects: Glucose Tolerance and Early-Phase Insulin Level Were Lower in the Biological Evening than in the Biological Morning.
The effects of circadian phase on glucose tolerance and insulin responses to test meals are shown in Figs. 3 and 4 and SI Appendix, Figs. S2–S5. Two-hour postprandial glucose AUC was 12% higher in the biological evening than in the biological morning (P < 0.0001), reflecting relatively reduced glucose tolerance in the biological evening. This effect depended on circadian misalignment exposure duration (P = 0.006). The difference between the biological evening and morning was more than twice as large on test day 1 (17%; P < 0.0001) than on test day 3 (7%; P = 0.0001). Similar results were found for peak postprandial glucose. This reduction in effect across test days may reflect a blunting of responses following repeated exposure to misalignment and/or a slight delay of the central endogenous circadian clock (as described earlier) as a result of continued light exposure during the biological night while misaligned. The effect on glucose tolerance of circadian phase (17%) was more than three times as large as the effect of the behavioral cycle (5%; P < 0.0001, paired t test) on test day 1, i.e., before central circadian phase had shifted substantially. This explains why the difference in glucose tolerance between breakfast and dinner while aligned was almost completely inverted during circadian misalignment on test day 1 (SI Appendix, Fig. S5). Early-phase postprandial insulin AUC was 27% lower in the biological evening than in the biological morning (P < 0.0001). Similar results were found for early-phase postprandial ISR AUC. Late-phase postprandial insulin AUC was not affected by circadian phase (P = 0.26), whereas late-phase postprandial ISR AUC was 8% higher in the biological evening than in the biological morning (P = 0.002). The effect of circadian phase on postprandial insulin and ISR was not dependent on circadian misalignment exposure duration (all P ≥ 0.066).
No Effect of Circadian Phase on Fasting Glucose, Yet Fasting Insulin Was Lower in the Biological Evening than in the Biological Morning.
The results for the effect of circadian phase on fasting glucose, insulin, and ISR are shown in SI Appendix, Figs. S3 and S6. There was no effect of circadian phase on fasting glucose (P = 0.43). Fasting insulin was 21% lower in the biological evening than in the biological morning (P = 0.024). Similar results were found for fasting ISR. Duration of exposure to circadian misalignment did not change these effects (all P ≥ 0.70).
Circadian Misalignment Reduced Glucose Tolerance Independent of Circadian Phase or Behavioral Effects.
The circadian misalignment results for glucose tolerance and insulin responses to tests meals are shown in Figs. 3 and 4 and SI Appendix, Figs. S2–S5. Two-hour postprandial glucose AUC was 6% higher during circadian misalignment than alignment (P = 0.0003), reflecting relatively lower glucose tolerance during misalignment. Similar results were found for peak postprandial glucose. Early-phase postprandial insulin AUC was not affected by circadian misalignment (P = 0.54). Similar results were found for early-phase postprandial ISR. Late-phase postprandial insulin AUC was 14% higher during circadian misalignment than alignment (P = 0.006), suggesting reduced insulin sensitivity during misalignment (as evidenced by misaligned postprandial glucose being higher despite late-phase postprandial insulin also being higher). Similar results were found for late-phase postprandial ISR. The effects of circadian misalignment on postprandial glucose, insulin, and ISR were sustained across the three consecutive days of circadian misalignment (all P ≥ 0.38).
No Effect of Circadian Misalignment on Fasting Glucose or Insulin.
The circadian misalignment results for fasting glucose, insulin, and ISR are shown in SI Appendix, Figs. S3 and S6. Circadian misalignment had no effect on fasting glucose, insulin, or ISR (all P ≥ 0.18). Duration of exposure to circadian misalignment did not change these effects (all P ≥ 0.33).
Circadian Misalignment Increased 24-h Glucose and Insulin Levels.
The results for 24-h glucose and insulin are shown in Fig. 5. Circadian misalignment increased 24-h glucose and insulin AUCs by 1.6% and 9%, respectively (both P ≤ 0.013). These effects depended on circadian misalignment exposure duration (both P = 0.002). When tested separately for each test day (and after Bonferroni adjustment), circadian misalignment had no statistical effect on 24-h glucose AUC for test day 1 (P = 0.052) or test day 3 (P = 0.094). Circadian misalignment had no effect on 24-h insulin AUC on test day 1 (P = 0.18), but significantly increased 24-h insulin AUC by 13% on test day 3 (P = 0.001).
Effects of circadian misalignment on 24-h glucose and insulin levels. TD, test day; gray bar represents sleep opportunity; black bar represents a meal. Probability values from 24-h AUC analyses are shown. Data are presented as mean ± SEM.
Postprandial Free Fatty Acid Levels Were Higher at Dinner Time, Independent of Circadian Effects, and Fasting Free Fatty Acid Levels Were Elevated at Dinner Time and by Circadian Misalignment.
The results for free fatty acid (FFA) levels are shown in SI Appendix, Figs. S7 and S8. Two-hour postprandial FFA AUC was 90% higher at dinner time than at breakfast time (P < 0.0001), but not different between circadian misalignment and alignment conditions (P = 0.059) or between the biological evening and morning (P = 0.11). Fasting FFA was 87% higher before dinner time than before breakfast time (P < 0.0001) and 15% higher during circadian misalignment than alignment (P = 0.018), but not different between the biological evening and the morning (P = 0.72). These effects were not dependent on circadian misalignment exposure duration (all P ≥ 0.16).
Circadian Misalignment Had No Significant Effect on Average 24-h FFA or Triglyceride Levels.
The results for 24-h FFA and triglyceride profiles are shown in Fig. 6. There was no effect of circadian misalignment on 24-h AUCs of FFA or triglyceride (both P ≥ 0.57), and this was not dependent on circadian misalignment exposure duration (both P ≥ 0.46). The FFA profiles were dependent on alignment condition (P < 0.0001), with higher levels at the start of the sleep opportunity and wake period, lower levels around lunch time, and higher levels at the end of the wake period in the circadian misalignment than alignment condition. The triglyceride profiles were also dependent on alignment condition (P < 0.0001), with lower levels during the latter portion of the sleep opportunity and for the first few hours after lights on, but higher levels in the latter part of the wake period in the circadian misalignment than alignment condition. These profile differences between alignment conditions were not dependent on circadian misalignment exposure duration (both P ≥ 0.46).
Effects of circadian misalignment on 24-h FFA and triglyceride levels. TD, test day; gray bar represents sleep opportunity; black bar represents a meal. Probability values from 24-h AUC analyses are shown. Data are presented as mean ± SEM.
Postprandial Respiratory Quotient and Carbohydrate Oxidation Rate Were Lower During Dinner and in the Biological Evening, and Reduced by Repeated Circadian Misalignment Exposure.
The results for postprandial respiratory quotient (RQ) and carbohydrate and lipid oxidation rates are shown in SI Appendix, Figs. S9 and S10. The behavioral cycle influenced postprandial substrate utilization, with RQ being 2% lower, carbohydrate oxidation rate being 7% lower, and lipid oxidation rate being 34% higher at dinner time compared with at breakfast time (all P ≤ 0.017). These findings were not dependent on circadian misalignment exposure duration (all P ≥ 0.45). Circadian phase also affected postprandial substrate use, with RQ being 2% lower and carbohydrate oxidation rate being 10% lower in the biological evening than morning (both P ≤ 0.033). There was no effect of circadian phase for postprandial lipid oxidation rate (P = 0.14). Duration of exposure to circadian misalignment influenced the effect of circadian phase on postprandial RQ and carbohydrate oxidation rate (both P ≤ 0.027) but not lipid oxidation rate (P = 0.085). Postprandial RQ and carbohydrate oxidation rate were lower in the biological evening than in the biological morning in the first test day (−4% and −19%, respectively; both P ≤ 0.003), but, by the third test day, there were no differences (both P ≥ 0.78). Circadian misalignment also impacted postprandial substrate use, reducing RQ by 2% and increasing lipid oxidation rate by 22% (both P ≤ 0.026). There was no overall impact of circadian misalignment on postprandial carbohydrate oxidation rate (P = 0.19). The effect of circadian misalignment on postprandial RQ and carbohydrate and lipid oxidation rates was dependent on exposure duration (all P ≤ 0.049). Circadian misalignment reduced postprandial RQ and carbohydrate oxidation rate by 3–11% on test day 3 (both P ≤ 0.004), without differences on test day 1 (both P ≥ 0.33). Circadian misalignment increased postprandial lipid oxidation rate by 50% on test day 3 (P = 0.004), without effect on test day 1 (P = 0.083).
Fasting RQ and Carbohydrate Oxidation Rate Were Lower Before Dinner, and Reduced by Circadian Misalignment.
The results for fasting RQ and carbohydrate and lipid oxidation rates are shown in SI Appendix, Fig. S11. The behavioral cycle influenced fasting substrate utilization, with RQ being 6% lower, carbohydrate oxidation rate being 31% lower, and lipid oxidation rate being 50% higher before dinner compared with before breakfast (all P < 0.0001). Circadian misalignment decreased fasted RQ (−3%) and fasted carbohydrate oxidation rate (−14%) and increased fasted lipid oxidation rate (+19%) compared with circadian alignment (all P ≤ 0.003). The effects of the behavioral cycle and circadian misalignment on fasted RQ and substrate oxidation rates were not dependent on circadian misalignment exposure duration (all P ≥ 0.30). Circadian phase had no effect on fasted RQ, or the fasted rates of carbohydrate or lipid oxidation (all P ≥ 0.058). However, the effects of circadian phase on RQ and lipid oxidation rate were dependent on circadian misalignment exposure duration (both P ≤ 0.038). Fasted RQ was 4% lower in the biological evening than the biological morning on the first test day (P = 0.006), but by the third test day there was no difference (P = 0.78). Similarly, fasted lipid oxidation rate was 26% higher in the biological evening than in the biological morning on the first test day (P = 0.006), but rates were not different by the third test day (P = 0.90). The effect of circadian phase on fasted carbohydrate oxidation rate was not dependent on circadian misalignment exposure duration (P = 0.081).
Circadian Misalignment Had No Effect on Average 24-h Growth Hormone Levels.
The results for 24-h growth hormone profiles are shown in SI Appendix, Fig. S12. Circadian misalignment had no effect on 24-h average growth hormone concentrations (P = 0.74), and this was not dependent on circadian misalignment exposure duration (P = 0.16). However, the shapes of the growth hormone profiles were dependent on alignment condition (P = 0.0001), with lower levels during the first few hours of the sleep opportunity and higher levels at the start of the wake period in the circadian misalignment condition compared with the circadian alignment condition. The profile difference between alignment conditions was not dependent on circadian misalignment exposure duration (P = 0.58).
Circadian Misalignment Decreased Total Sleep Time (TST).
The polysomnography results are shown in SI Appendix, Figs. S13 and S14. Circadian misalignment, compared with circadian alignment, decreased TST by 56 min (P < 0.0001). This effect occurred despite misalignment related reductions of 15 and 7 min in latencies to both N1 and N2 sleep, respectively (both P ≤ 0.004). Circadian misalignment also decreased the durations of N2 and rapid eye movement (REM) sleep by 41 and 24 min, respectively (both P < 0.0001), but did not significantly affect the amounts of N1 or N3 sleep (both P ≥ 0.098). The circadian misalignment effects on TST and REM sleep were dependent on the duration of exposure to circadian misalignment (both P ≤ 0.028). TST was 72 min shorter during circadian misalignment than alignment on test day 1 (P < 0.0001), and the misalignment-induced reduction in TST was only 42 min by test day 3 (P = 0.001). REM sleep duration was 36 min shorter in the circadian misalignment than alignment protocol on test day 1 (P < 0.0001), but there was no difference on test day 3 (P = 0.071). The increases in TST and REM sleep over repeated days of circadian misalignment may be because the circadian system was slightly more appropriately aligned with the daytime sleep opportunity on the third test day than on the first, because the influence of the circadian system was diminished on the third test day vs. the first (as suggested by the blunting of the cortisol and melatonin profiles), and/or because homeostatic sleep pressure was increased following repeated bouts of impaired sleep. The effects of circadian misalignment did not significantly depend on circadian misalignment exposure duration for N1, N2, or N3 sleep durations, or latencies to N1 and N2 sleep (all P ≥ 0.15).
Relationships Between Glucose Metabolism and Sleep Measured by Polysomnography.
In covariance analyses, we tested whether sleep parameters significantly explained variance in glucose and insulin AUCs. TST and all sleep stages were nonsignificant covariates (all P ≥ 0.055) in our 2-h postprandial glucose AUC and early- and late-phase postprandial insulin AUC analyses.
Discussion
Our results revealed separate effects of the endogenous circadian system and of circadian misalignment, independent from effects of the behavioral cycle, on glucose tolerance in humans. Glucose tolerance was 17% lower in the biological evening than in the biological morning on test day 1 (before the central circadian pacemaker had substantially shifted), independent of the behavioral cycle, and 6% lower when the behavioral cycle was inverted relative to the circadian system, similar to the conditions experienced by night workers. Glucose tolerance was also 8% lower at dinner time than at breakfast time, independent of endogenous circadian phase. Under normally entrained conditions, i.e., when sleep occurs at night and nutrients are consumed during the day, glucose tolerance has been shown to deteriorate from the morning to the evening in healthy individuals (1⇓⇓⇓⇓–6). Here we show that there are two separate contributing factors to this deterioration: the deterioration from breakfast to dinner (the behavioral cycle effect, independent of circadian phase) and the deterioration from the biological morning to the biological evening (the endogenous circadian cycle effect, independent of the behavioral cycle). Our findings suggest that the circadian system per se strongly affects glucose tolerance and thereby importantly affects 24-h glucose regulation. On test day 1, the impact of the circadian system on glucose tolerance was more than three times the magnitude of the difference between the first and last meal of the day (comparing responses to test meals consumed at breakfast time vs. dinner time, regardless of circadian phase). In addition, our results suggest that circadian misalignment per se lowers glucose tolerance and thus could increase diabetes risk, which has particular relevance to shift workers. We also found that the effect of the circadian system and circadian misalignment on glucose tolerance could be mediated, at least in part, by two different insulin mechanisms: (i) lower glucose tolerance in the biological evening was related to a 27% lower early-phase insulin response, implying reduced β-cell function; and (ii) lower glucose tolerance during circadian misalignment was associated with a 14% higher late-phase insulin response despite elevated postprandial glucose concentrations—suggesting decreased insulin sensitivity. Other mechanisms could also contribute to the separate circadian phase and circadian misalignment effects on glucose tolerance. These include, but are not limited to, differences in gastrointestinal absorption, hepatic glucose output suppression, and non–insulin-dependent glucose metabolic pathways.
Glucose tolerance was lower in the biological evening than the biological morning (independent of the behavioral cycle), and this effect was associated with reduced early-phase insulin secretion—indicative of an insufficient β-cell response. These results are in line with prior research showing circadian control of glucose metabolism in humans (4, 10⇓–12, 14). Our finding of a circadian phase effect on early-phase insulin secretion could be mediated through the circadian system via numerous mechanisms. First, there are multisynaptic projections from the SCN to the pancreas (33, 34). Second, pancreatic cells contain circadian clocks and their disruption, by knocking out circadian clock gene Brain and muscle Arnt-like protein-1 (BMAL1) specifically within the pancreas, results in decreased insulin secretion and impaired glucose tolerance (21, 23). Third, endogenous circadian rhythms in hormones controlled by the SCN (e.g., corticosteroids and melatonin) may help entrain peripheral clocks, including those within pancreatic islets (35, 36). Fourth, endogenous circadian hormone rhythms (e.g., melatonin) can directly influence human β-cell compensation (37⇓–39).
In addition to circadian modulation of β-cell function, the circadian system may also influence glucose tolerance through insulin sensitivity and hepatic glucose output via similar pathways as discussed for the pancreas. For example, there is evidence for the autonomic nervous system controlling 24-h variation in circulating glucose levels by affecting the liver. Hepatic sympathectomy results in the disruption of the 24-h rhythm in plasma glucose levels in rats (40), similar to that observed following SCN lesions (24).
The circadian rhythm in melatonin is unlikely to explain the lower glucose tolerance and β-cell function in the biological evening in the present study because melatonin levels were actually lower immediately preceding the test meals in the biological evening compared with the biological morning. The circadian rhythm in cortisol is also unlikely to explain the lower glucose tolerance in the biological evening because cortisol, which decreases glucose tolerance (41, 42), actually peaks in the biological morning and is low in the biological evening (10, 15). Raised FFA levels can impair glucose tolerance and decrease insulin sensitivity and glucose oxidation (43⇓–45). However, we found no effect of circadian phase on fasting or postprandial FFA levels. We were unable to directly assess with i.v. glucose tolerance tests or hyperinsulinemic-euglycemic clamps if circadian phase affects insulin sensitivity. However, there is strong evidence from rodent studies that the circadian system influences insulin sensitivity (19, 20, 46). We found that postprandial carbohydrate oxidation was lower in the biological evening than in the biological morning, similar to the difference in glucose tolerance. Decreases in postprandial carbohydrate oxidation have been observed in people with impaired glucose tolerance, such as those with type 2 diabetes (47). Although we assessed two primary hormones regulated by the circadian system, i.e., melatonin and cortisol, future studies are required to assess other neuroendocrine mechanisms underlying the effect of the human circadian system on glucose tolerance.
The circadian phase effect on glucose tolerance was blunted following consecutive days of circadian misalignment. The attenuation in the biological morning/evening glucose tolerance difference following successive days of circadian misalignment may be attributed to blunted, shifted, or otherwise disrupted circadian function. Animal experimental studies have shown that prolonged misalignment between the circadian system and the 24-h environmental cycle (e.g., through continuous light exposure or simulated shift work protocols) blunts and/or phase-shifts many circadian rhythms, including those in behavior, endocrinology, and clock gene expression, and causes desynchrony between individual oscillators (26⇓–28, 48, 49). We had no direct measure of central or peripheral circadian clock function changes across repeated days of circadian misalignment. However, we did obtain evidence suggestive of a blunting and/or a phase shift of the rhythms in circulating melatonin and cortisol concentrations across repeated days of circadian misalignment, suggesting that rhythmic output of the central circadian pacemaker was blunted and/or phase-shifted. It is known from animal experiments that peripheral circadian clocks in organs related to metabolic function (e.g., in the liver and pancreas) are rapidly entrained to a reversed feeding schedule (i.e., within a few days), whereas the central circadian pacemaker is not (50, 51). The resultant internal desynchrony between the central and peripheral clocks has been proposed to underlie adverse metabolic consequences of shift work exposure (52).
We have previously shown that circadian misalignment—while living on a 28-h behavioral cycle under dim light conditions (i.e., a forced desynchrony protocol)—resulted in elevated postprandial glucose concentrations (10). However, real-life shift workers do not live on 28-h days in dim light. Thus, to assess the effects of night work under more realistic conditions, in the present protocol, individuals lived on 24-h days and were exposed to typical room light intensity during their wake episodes. The present study showed that the adverse effects of circadian misalignment during a forced desynchrony protocol are also observed in conditions more similar to those experienced by real-life shift workers, and are sustained over a number of days of repeated exposure. Our data are also consistent with the general notion that various forms of circadian disruption (e.g., SCN lesions, clock gene mutations, continuous light exposure, and simulated shift work) can impair glucose metabolism (19, 20, 26, 28, 29).
The circadian misalignment effect on glucose tolerance in the present study appeared to be mediated mostly by a decrease in insulin sensitivity, rather than a decrease in β-cell function—although β-cell compensation was also inadequate. In our circadian misalignment protocol, growth hormone and fasting FFA levels, which can decrease insulin sensitivity (43, 53), were increased during the wake episodes and thereby could help explain why circadian misalignment decreased our estimate of insulin sensitivity. In addition, there was a statistical trend for circadian misalignment increasing postprandial FFA levels. In humans, daytime pharmacological melatonin administration impairs insulin secretion, decreases insulin sensitivity, and reduces glucose tolerance (37, 39). In the present study, the melatonin rhythm, albeit blunted, was reversed by circadian misalignment, with levels peaking during the wake period (when participants ate) rather than during the sleep period. If physiological concentrations of melatonin have a similar effect, this could have contributed to reduced insulin sensitivity and insufficient β-cell response to the meals and resulted in lowered glucose tolerance during circadian misalignment. Decreased insulin sensitivity is a key early defect in type 2 diabetes development; therefore, our findings and prior reports of decreased insulin sensitivity with circadian disruption (caused by lesioning the SCN or knocking out clock genes) provide a possible explanation for why shift workers are more likely to develop impaired glucose metabolism and type 2 diabetes (7⇓–9, 19, 20, 54). As stated earlier, food intake strongly entrains peripheral clocks, but not the central clock, in rodents, although it is unknown if this effect occurs in humans (51). Thus, the nighttime food intake in our circadian misalignment protocol may have caused internal desynchrony and thereby conflicting signals in the regulation of metabolism from central vs. peripheral clocks (e.g., within the liver and pancreas). Such internal desynchrony, if it occurred, may have contributed to the adverse effects of circadian misalignment on glucose metabolism in the present study. We found no evidence for short-term circadian misalignment reducing β-cell function, but other studies have shown that prolonged clock gene disruption and desynchrony between the circadian system and 24-h environmental/behavioral cycles can impair β-cell function (18, 21, 28, 32, 55).
The behavioral cycle also affected FFA levels, with premeal and postprandial levels being higher at dinner time than at breakfast time, independent of circadian phase. As stated earlier, FFAs can impair glucose tolerance, decrease insulin sensitivity, and reduce glucose oxidation (43⇓–45). Thus, the higher fasting and postprandial FFA levels before and after dinner could help explain why glucose tolerance and our estimate of insulin sensitivity were lower at dinner time. Cortisol can decrease glucose uptake and decrease insulin sensitivity (41, 42). There is a behavioral cycle influence on cortisol (although the effect is much smaller than the endogenous circadian system’s influence), with levels being lower before dinner than breakfast (10, 15). Indeed, in the present study, cortisol levels were 46% lower before dinner than before breakfast, independent of circadian phase (P < 0.0001). Thus, the behavioral cycle-driven rhythm in cortisol is unlikely to contribute to the behavioral cycle’s effect on glucose tolerance. Further research is required to determine which other regulators of glucose metabolism are affected by the behavioral cycle.
Experimental sleep restriction and slow-wave sleep suppression reduce glucose tolerance, decrease β-cell function, and decrease insulin sensitivity (56⇓–58). In our circadian misalignment protocol, sleep duration was decreased and the durations of sleep stages were altered [although slow-wave sleep (N3) was not changed]. No sleep parameters significantly explained variance in our postprandial glucose and insulin models. A recent study demonstrated that circadian misalignment decreases insulin sensitivity, independent of sleep loss (31).
Strengths of the present study include that all measurements were conducted under the same behavioral and environmental conditions, including semirecumbent posture, physical inactivity, and 90-lux illuminance for the full duration of each identical test meal session. The study design allowed the separate assessment of the influence of the behavioral cycle (breakfast vs. dinner), circadian phase (8:00 AM vs. 8:00 PM), and circadian alignment vs. circadian misalignment, as well as the assessment of repeated exposure to circadian alignment and misalignment. Limitations of the present study also need to be considered. First, we only assessed postprandial glucose and insulin responses at two behavioral and circadian cycle phases. Even though these were targeting conditions when we expect large differences, the differences between these conditions could underestimate the maximal effect of the behavioral cycle and/or the circadian system if the peaks and/or troughs of the behavioral cycle and circadian system effects were missed. Test meal assessments occurred at 8:00 AM and 8:00 PM for several reasons: (i) these times are compatible with a typical meal schedule without interrupting sleep; (ii) they target the maximum and minimum glucose tolerances observed under normal behavioral schedules (i.e., not eating during the night when one would normally sleep); (iii) glucose tolerance following meals can be assessed only after at least ∼8 h of fasting, thereby preventing us assessing it more than twice per day; and (iv) by scheduling meals 12 h apart, they occurred at the same clock time in the aligned and misaligned conditions. Second, we could not determine the relative contributions of different components of the behavioral cycle (e.g., sleep/wake cycle vs. fasting/feeding cycle) on glucose metabolism. We purposely designed the present study such that there was a shorter fasting period before dinner than breakfast because this is typically the case for most meal schedules, and this difference in fasting duration probably contributed to the effect of the behavioral cycle on glucose metabolism. Third, we assessed glucose and insulin responses to test meals. To further test whether changes in insulin sensitivity vs. insulin secretion explain the differences in glucose tolerance, one may consider using techniques such as i.v. glucose tolerance tests and hyperinsulinemic-euglycemic clamps, yet the use of such methods twice per wake period would severely disrupt the fasting/feeding cycle and physiology in ways that substantially impact subsequent assessments, and thereby affect the ability to test the separate effects of the behavioral cycle, circadian phase, and circadian misalignment on glucose metabolism. Fourth, our circadian misalignment exposure lasted only a few days. The effect of prolonged circadian misalignment per se on glucose metabolism is unknown. Fifth, our participants were healthy and had no significant shift work experience. The effects of the behavioral cycle, circadian phase, and circadian misalignment on glucose metabolism may be different in people with diabetes and in shift workers.
We have systematically investigated the separate effects of the behavioral cycle, circadian phase, and circadian misalignment on glucose tolerance and insulin responses in healthy humans. Glucose tolerance was lower in the biological evening than in the biological morning (irrespective of the behavioral cycle), and was reduced by circadian misalignment, independent of circadian phase and behavioral effects. These two effects on glucose tolerance were seemingly mediated, at least in part, by two different insulin-related mechanisms. Furthermore, our findings indicate that two separate processes contribute to the typical decrease in glucose tolerance observed in healthy individuals from morning to evening, namely the behavioral cycle plus a critical contribution from the circadian system. Further, our findings show that, independent from these effects of the circadian system and the behavioral cycle, circadian misalignment itself lowers glucose tolerance, without diminishing or worsening effects upon repeated daily exposures to circadian misalignment. Thus, these findings have implications for glucose regulation and may help explain why shift work is a risk factor for type 2 diabetes. Our observations underscore results from recent studies in humans and animals that suggest that it is not merely what we eat, but also when we eat, that has important health consequences, including for glucose metabolism (59). Our findings may also help the development of behavioral and circadian strategies (e.g., timing of eating) that could improve glycemic control in day-active people and night workers. More research is needed to determine if the effects of the behavioral cycle, circadian phase, and circadian misalignment are altered in populations with impaired glucose tolerance or type 2 diabetes and in shift workers.
Materials and Methods
Experimental Design.
Each participant underwent two 8-d laboratory protocols, according to a cross-over design, to test the separate effects of the behavioral cycle, circadian phase, and circadian misalignment on glucose metabolism (Fig. 2). One protocol included circadian misalignment and the other maintained circadian alignment. The visits were separated by 2–8 wk (mean ± SD, 4 ± 2 wk). “Minimization” was used to minimize imbalance—according to age, sex, and body mass index (BMI)—in the order of laboratory visits (seven participants undertook the circadian alignment protocol first, the other seven participants undertook the circadian misalignment protocol first) (60).
Participants.
Fourteen healthy nonsmoking, drug- and medication-free (except for oral contraceptive agents) adults completed this study [mean age ± SD (range), 28 ± 9 y (20–49 y); BMI, 25.4 ± 2.6 kg/m2 (21–29.5 kg/m2); eight men]. Health status was determined by physical examination, standard laboratory tests, and psychiatric assessment. Participants reported no shift work in the past 3 y and less than 6 mo cumulative lifetime shift work exposure and had not crossed more than one time zone in the previous 3 mo. Participants provided written informed consent. The Partners Human Research Committee approved this research, which was conducted in the Center for Clinical Investigation (CCI) at Brigham and Women’s Hospital (Boston, MA).
Preinpatient Study Conditions.
Participants selected and maintained a normal sleep/wake schedule, with an 8-h sleep opportunity, for ≥11 d (mean ± SD, 17 ± 3 d) before each laboratory visit. Participants were instructed to sleep between 11:00 PM and 7:00 AM on the night preceding each inpatient admission to aid the adaptation of the participant’s endogenous circadian system to the initial laboratory sleep/wake schedule (sleep opportunity, 11:00 PM to 7:00 AM). Compliance was assessed with wrist actigraphy [Actiwatch Spectrum (Philips-Respironics) or Actiwatch-L (Mini Mitter)], sleep diary, and daily bedtime and wake time calls to a time-stamped voicemail system (mean ± SD, bedtime, 11:30 PM ± 48 min; wake time, 7:21 AM ± 41 min; data from seven sleep periods preceding the final ambulatory sleep period before both inpatient admissions).
Inpatient Study Conditions.
On the first day of each 8-d laboratory protocol, participants were admitted to the CCI at ∼10:30 AM to undertake the circadian alignment protocol or circadian misalignment protocol, in a cross-over design (Fig. 2). Participants remained in a private laboratory room throughout each laboratory protocol to allow strict control of environmental conditions. In the circadian alignment protocol, the participant’s sleep opportunity occurred between 11:00 PM and 7:00 AM for days 1–8. In the circadian misalignment protocol, the participant’s sleep opportunity occurred between 11:00 PM and 7:00 AM for days 1–3. On day 4 of the circadian misalignment protocol, the participant’s behavioral cycles were shifted by 12 h, and this was maintained until the end of that protocol (day 8). The 12-h shift on day 4 was achieved by including an 8-h wake episode and a 4-h sleep opportunity, thereby maintaining the same sleep opportunity-to-wake ratio (1:2) in the circadian alignment and misalignment protocols. Metabolic responses to test meals were assessed on days 5 and 7 in the circadian alignment protocol and across days 5/6 and 7/8 in the circadian misalignment protocol (as detailed later). Light levels—in the horizontal angle of gaze—during the protocols are shown in Fig. 2: ∼90 lux to simulate typical room light intensity, ∼450 lux during the first three baseline wake episodes to enhance circadian entrainment, 30-min periods of ∼450 lux to simulate the morning commute preceding the day work shift (circadian alignment protocol) and following the night work shift (circadian misalignment protocol), ∼4 lux to permit assessment of the dim-light melatonin onset, and 0 lux during scheduled sleep episodes.
Diet.
Participants were given an ad libitum lunch at approximately 12:00 PM on the first day of each laboratory protocol. Thereafter, participants received an isocaloric diet, calculated according to the Harris–Benedict equation with an activity factor of 1.4. The diet consisted of 45–50% carbohydrate, 30–35% fat, and 15–20% protein, with 150 mEq Na+ (±20%) and 100 mEq K+ (±20%), and at least 2.5 L of water per 24 h. Participants were instructed to consume all food provided (verified by checking their food trays). Diet was identical within each participant between laboratory visits, except for the required and prorated (50% of a 24-h cycle) additional food and water given during the 12-h behavioral cycle (day 4) in the circadian misalignment protocol.
We assessed participants’ metabolic responses to identical test meals (33.3% of calculated daily calorie intake) given 1 h and 13 h following scheduled wake time in the circadian alignment (wake periods 5 and 7) and misalignment (wake periods 6 and 8) protocols. Participants chose one of two test meals (i): a dextrose solution (Glucola, 0.45 g/kg), a bagel with butter, cereal with milk and sugar, egg, and peanuts; (ii) Glucola (0.45 g/kg), a bagel with butter, cereal with milk and sugar, turkey sausage, and almonds. Glucola was consumed within the first 1 min and other food items were consumed subsequently in the order listed. Test meals were consumed within 20 min and were identical within each participant across both protocols. These meals were preceded by isocaloric “pre-meals”: “dinner” (participants preselected one of two meals, which was identical within each participant across both protocols) preceded each test “breakfast” on the prior wake period (circadian alignment protocol, 8:00 PM; circadian misalignment protocol, 8:00 AM) and “lunch” (participants preselected one of two meals, which was identical within each participant across both protocols) preceded each test dinner on the metabolic test days (circadian alignment protocol, 11:30 AM; circadian misalignment protocol, 11:30 PM).
Blood Sampling During Metabolic Test Days.
The 24-h blood drawing for metabolite and hormone assessment (SI Appendix provides assay details) started shortly after bedtime until bedtime 24 h later, i.e., between 11:00 PM and 11:00 PM in the circadian alignment protocol (sleep period 4 and wake period 5 and sleep period 6 and wake period 7) and between 11:00 AM and 11:00 AM in the circadian misalignment protocol (sleep period 5 and wake period 6 and sleep period 7 and wake period 8). Blood drawing difficulties during sleep opportunities precluded the use of all subjects in the 24-h glucose (n = 10), insulin (n = 10), FFA (n = 11), and triglyceride (n = 11) AUC analyses. For each of the eight test meals per person, fasting blood was drawn 7 min before the meal, and postprandial blood was drawn every 10 min for 90 min, starting 10 min after the participant began eating the test meal, and every 30 min for the next 90 min, totaling 3 h.
Indirect Calorimetry.
Indirect calorimetry measurements for test meals sessions were obtained in 11 subjects with a calibrated, open-circuit, ventilated hood system (Vmax Encore 29N; VIASYS Healthcare; detailed in SI Appendix). Technical difficulties precluded indirect calorimetry measurements in 3 of the 14 participants.
Polysomnography.
Sleep was recorded by polysomnography (Vitaport; TEMEC Instruments)—in accordance with the American Academy of Sleep Medicine recommendations (61)—during sleep periods 1, 4, and 6 in the circadian alignment protocol and during sleep periods 1, 5, and 7 in the circadian misalignment protocol (detailed in SI Appendix).
Data Analysis and Statistics.
ISR was estimated from serum C-peptide levels (62). The calculations were performed by computer program “Insulin SECretion” (63) that was kindly provided by Roman Hovorka (University of Cambridge, Cambridge, United Kingdom). Postprandial glucose AUC was calculated from fasting to 120 min relative to the start of the meal. For test meal analysis of insulin and ISR, early-phase responses were defined as AUC from fasting to 30 min following the start of the meal, and late-phase response as the AUC between 30 and 120 min relative to the start of the meal. AUC was calculated by using the trapezoidal method. Peak glucose concentration was determined during the 2-h postprandial period.
Unless otherwise stated, statistical tests were performed with linear mixed models, with participant included as a random factor. Where necessary, analysis was performed on log-transformed data. Statistical significance was accepted as P < 0.05. Data are presented as mean ± SEM unless otherwise indicated. SI Appendix provides further details.
Acknowledgments
We thank the research volunteers and Center for Clinical Investigation nursing and technical staff. We also thank Janis F. Swain, RD; Karen Yee, RD; and Leigh K. Keating, RD, for their expert assistance with diet preparation. This study was supported by National Heart, Lung, and Blood Institute (NHLBI) Grant R01 HL094806 (to F.A.J.L.S.). C.J.M. was partly supported by the National Space Biomedical Research Institute through National Aeronautics and Space Administration Grant NCC 9-58. O.M.B. was partly supported by National Institute on Aging Grant P01 AG009975. I.B. was supported by the Brazilian National Council for Scientific and Technological Development. F.A.J.L.S. was supported in part by NHLBI Grant R01 HL094806, the Fund to Sustain Research Excellence by the Brigham and Women's Hospital Biomedical Research Institute, National Institute of Diabetes and Digestive and Kidney Diseases Grant R01 DK099512, and NHLBI Grant R01 HL118601. S.A.S. was supported in part by NHLBI Grant K24 HL076446. This project was supported by Clinical Translational Science Award UL1RR025758 to Harvard University and Brigham and Women’s Hospital from the National Center for Research Resources.
Footnotes
- ↵1To whom correspondence may be addressed. Email: cjmorris{at}partners.org or fscheer{at}rics.bwh.harvard.edu.
Author contributions: C.J.M. and F.A.J.L.S. designed research; C.J.M., J.N.Y., J.I.G., S.M., I.B., and F.A.J.L.S. performed research; C.J.M., W.W., and F.A.J.L.S. analyzed data; and C.J.M., O.M.B., S.A.S., and F.A.J.L.S. wrote the paper.
Conflict of interest statement: O.M.B. has received two investigator-initiated grants from Sepracor (now Sunovion; ESRC-0004 and ESRC-0977; ClinicalTrials.gov identifiers NCT00555750 and NCT00900159) and two investigator-initiated grants from Cephalon (now Teva; ClinicalTrials.gov identifier NCT00895570); received Speaker’s Bureau, continuing medical education (CME) and non-CME lecture honoraria, and an unrestricted educational grant from Takeda Pharmaceuticals North America; served as a consultant and expert witness for Dinsmore and received consulting fees for serving on the Scientific Advisory Board of Matsutani America and consulting fees from the Wake Forest University Medical Center; received speaking fees and/or travel support for speaking from American Academy of Craniofacial Pain, National Heart, Lung, and Blood Institute, National Institute of Diabetes and Digestive and Kidney Diseases, National Postdoctoral Association, Oklahoma State University, Oregon Health & Science University, State University of New York Downstate Medical Center, American Diabetes Association, and New York University.
This article is a PNAS Direct Submission.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1418955112/-/DCSupplemental.
References
- ↵.
- Saad A, et al.
- ↵.
- Carroll KF,
- Nestel PJ
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵.
- Gan Y, et al.
- ↵.
- Scheer FA,
- Hilton MF,
- Mantzoros CS,
- Shea SA
- ↵
- ↵
- ↵
- ↵.
- Frank SA, et al.
- ↵
- ↵
- ↵
- ↵.
- Turek FW, et al.
- ↵.
- Coomans CP, et al.
- ↵
- ↵
- ↵.
- Lamia KA,
- Storch KF,
- Weitz CJ
- ↵
- ↵
- ↵.
- Karatsoreos IN,
- Bhagat S,
- Bloss EB,
- Morrison JH,
- McEwen BS
- ↵.
- Fonken LK, et al.
- ↵
- ↵.
- Qian J,
- Block GD,
- Colwell CS,
- Matveyenko AV
- ↵.
- Hampton SM, et al.
- ↵
- ↵.
- Leproult R,
- Holmbäck U,
- Van Cauter E
- ↵.
- Buxton OM, et al.
- ↵
- ↵
- ↵
- ↵.
- Balsalobre A, et al.
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵.
- la Fleur SE,
- Kalsbeek A,
- Wortel J,
- Fekkes ML,
- Buijs RM
- ↵.
- Boden G,
- Ray TK,
- Smith RH,
- Owen OE
- ↵
- ↵
- ↵.
- Stokkan KA,
- Yamazaki S,
- Tei H,
- Sakaki Y,
- Menaker M
- ↵.
- Damiola F, et al.
- ↵
- ↵.
- Møller N, et al.
- ↵
- ↵.
- Gale JE, et al.
- ↵.
- Buxton OM, et al.
- ↵
- ↵.
- Tasali E,
- Leproult R,
- Ehrmann DA,
- Van Cauter E
- ↵.
- Mattson , et al.
- ↵.
- Altman DG,
- Bland JM
- ↵.
- Iber C,
- Ancoli-Israel S,
- Chesson A,
- Quan SF
- ↵.
- Van Cauter E,
- Mestrez F,
- Sturis J,
- Polonsky KS
- ↵
Citation Manager Formats
Article Classifications
- Biological Sciences
- Neuroscience