Prolonged, Controlled Daytime Versus Delayed Eating Impacts Weight and Metabolism

Summary

A delayed eating schedule is associated with increased risk of obesity and metabolic dysfunction in humans [1–9]. However, there are no prolonged, highly controlled experimental studies testing the effects of meal timing on weight and metabolism in adults with a body mass index (BMI) of 19-27 kg/m² [10–18]. Twelve healthy adults (age: 26.3±3.4y; BMI: 21.9±1.7 kg/m²; 5 females) participated in a randomized crossover study in free-living conditions. Three meals and two snacks with comparable energy and macronutrient contents were provided during two, 8-week, counterbalanced conditions separated by a 2-week washout period: 1) daytime (intake limited to 0800h-1900h); and 2) delayed (intake limited to 1200h-2300h). Sleep-wake cycles and exercise levels were held constant. Weight, adiposity, energy expenditure, and circadian profiles of hormones and metabolites were assessed during four inpatient visits occurring before and after each condition. Body weight, insulin resistance (HOMA-IR), trunk-to-leg fat ratio, resting energy expenditure, respiratory quotient, and fasting glucose, insulin, total and dHDL cholesterol, and adiponectin decreased on the daytime compared to the delayed schedule. These measures, as well as triglycerides increased on the delayed compared to the daytime schedule (effect size range: d=0.397-1.019). Circadian phase and amplitude of melatonin, cortisol, ghrelin, leptin, and glucose were not differentially altered by the eating schedules. Overall, an 8-week daytime eating schedule, compared to a delayed eating schedule, promotes weight loss and improvements in energy metabolism and insulin in adults with BMI 19-27 kg/m², underscoring the efficacy and feasibility of daytime eating as a behavioral modification for real world conditions.

eTOC Blurb

Allison et al. demonstrate that consuming meals earlier in the day promotes a healthier metabolic profile and weight than eating later in the day. This suggests that curtailing eating in the evening is a helpful strategy for managing cardiometabolic health and weight.

Results

Twenty-nine participants were screened, and 17 were enrolled. Of these 17 participants, four withdrew: two after the initial inpatient visit, citing study burden, and two during the first eating condition due to unexpected relocations. Thirteen participants completed the study, but one participant regularly did not consume most of the snacks provided on either condition in a deliberate effort to lose weight; as such, that participant’s data were excluded. This yielded a final sample of 12 participants (mean±SD age, 26.3±3.4y; BMI, 21.9±1.7 kg/m²; 5 females; 8 non-Hispanic white, 3 Asian, and 1 non-Hispanic black). The five excluded/non-completing participants (mean±SD age, 25.7±8.4y; BMI, 23.0±2.0 kg/m²; 4 females; 5 non-Hispanic white) did not differ from completers (p’s >0.16). Those randomized to the daytime versus the delayed condition first showed no sex, age, race, or BMI differences (p’s >0.19).

Compliance to Eating Conditions

Compliance to eating parameters, physical activity and sleep during each 8-week eating condition was excellent (Table 1). Only one sleep measure significantly differed between conditions: actigraphic sleep onset was 25 minutes later during the delayed than daytime condition (p=0.04). Self-reported physical activity, and actigraphic total activity counts and light, medium, and vigorous activity did not differ significantly between conditions (all p’s>0.05; data not shown for types of activity). Figure S1 shows activity counts across 24h in 10-minute intervals in the two conditions. White, blue, green, and red-light exposure, as measured by actigraphy, also did not significantly differ between conditions (all p’s>0.05; data not shown).

Table 1.

Compliance to Conditions for Timed Eating, Physical Activity and Sleep Measures

Outcome Measures	Daytime Mean (SD)	Delayed Mean (SD)	P-value
Timed Eating Compliance
# Meals or snacks/day	4.6 (0.6)	4.6 (0.6)	0.91
# Eating episodes out of range* (per 8-week condition)	5.4 (5.9)	2.2 (2.8)	0.052
# Eating episodes with non-study food items consumed	8.0 (5.5)	8.8 (7.8)	0.60
Calories provided (kcals)	2018.8 (289.1)	2001.6 (288.3)	0.88
Calories consumed/kg (kcals)	1998.7 (359.9)	2005.1 (347.1)	0.97
Physical Activity Compliance
Minutes/day (self-report)	20.0 (13.6)	24.9 (13.9)	0.056
Total activity counts/day (actigraphy)	250,664.8 (66,777.1)	267,566.9 (93,149.5)	0.17
Actigraphic Sleep Compliance
Sleep duration (h)	7.7 (0.6)	7.5 (0.6)	0.09
Sleep onset (hh:mm; h)	23:58 (0.7)	24:23 (0.9)	0.04
Sleep offset (hh:mm; h)	07:40 (0.9)	07:54 (1.0)	0.14
Sleep midpoint (hh:mm; h)	03:49 (0.7)	04:09 (0.9)	0.07
Sleep onset latency (min)	28.9 (15.0)	25.5 (14.4)	0.30
Sleep efficiency (%)	83.0 (3.9)	83.2 (4.6)	0.80
Wake after sleep onset (min)	38.7 (15.2)	39.7 (15.2)	0.41

Note: Paired t-tests compared compliance data.

^*A mean (SD) of 5.3 (5.8) episodes of the out of range eating during the 8-week daytime condition occurred after 1900h; a mean (SD) of 1.5 (2.0) episodes of the out of range eating during the 8-week delayed condition occurred before 1200h. See also Figure S1.

Comparison of Changes in Body Weight, Energy Expenditure, Hormones, and Other Metabolites

Body weight decreased during daytime eating, and increased during delayed eating, with a medium effect size (Table 2, Figure 1A). Resting energy expenditure (REE) and respiratory quotient (RQ; i.e., VCO₂/VO₂) decreased with medium and small effect sizes, respectively, during daytime eating and increased during delayed eating, the latter indicating decreased fat oxidation in the delayed condition (Table 2, Figure 1B-3C). Body composition measured by dual x-ray absorptiometry (DXA) revealed a small decrease in % total body fat in the delayed condition and a small increase in the daytime condition. Lean mass % did not change appreciably in either condition. By contrast, the % trunk fat decreased with a small effect size in the daytime condition. The trunk-to-leg fat ratio decreased during daytime compared to delayed eating, with a large effect size and significant difference (Table 2, Figure 1D). Baseline measures before each condition did not differ significantly (data not shown).

Table 2.

Comparison of Change Values in Weight, Metabolic Markers, and Fasting Hormone and Metabolites Between Conditions

	Daytime Condition		Delayed Condition		Analyses
					t-test		Effect Size
	Pre-	Post-	Pre-	Post-
OUTCOME MEASURES	Mean (SD)		Mean (SD)		t	p	d
Weight and Metabolic Markers
Weight (kg)	65.4 (7.5)	64.3 (7.6)	64.8 (7.5)	65.0 (8.0)	−1.976	0.074	0.571
Resting Energy Expenditure (kcals)	1570.9 (214.5)	1477.3 (230.1)	1563.2 (214.5)	1587.6 (138.0)	−1.783	0.100	0.515
Respiratory Quotient (CO2 eliminated/O2 consumed)	0.84 (0.05)	0.83 (0.04)	0.81 (0.06)	0.84 (0.07)	−1.571	0.140	0.454
DXA Total Fat (%)	27.6 (5.4)	27.9 (5.0)	28.0 (5.4)	27.6 (5.0)	1.647	0.130	0.476
DXA Lean Mass (%)	14.5 (2.0)	14.7 (1.9)	14.7 (1.9)	15.0 (2.2)	−0.673	0.510	0.194
DXA Trunk Fat (%)	26.1 (5.3)	25.7 (5.1)	25.7 (5.3)	25.6 (5.1)	−1.068	0.310	0.308
DXA Trunk/Leg Fat (%)	0.87 (0.11)	0.85 (0.10)	0.85 (0.11)	0.87 (0.11)	−3.530	0.005	1.019
Fasting Hormones and Metabolites*
Total Cholesterol (mg/dL)	159.8 (29.1)	155.9 (28.2)	157.6 (38.0)	162.0 (34.2)	−1.416	0.184	0.409
dHDL Cholesterol (mg/dL)	57.4 (12.6)	53.8 (9.1)	56.3 (15.5)	60.0 (16.8)	−2.536	0.028	0.732
LDL Cholesterol (mg/dL)	87.9 (24.6)	87.8 (25.7)	86.2 (32.6)	83.7 (31.2)	1.857	0.090	0.536
Triglycerides (mg/dL)	71.8 (19.2)	73.3 (32.7)	78.4 (12.6)	93.4 (46.9)	−1.895	0.085	0.547
Adiponectin (ug/ml)	17.7 (8.3)	17.4 (6.3)	17.4 (6.7)	21.4 (6.6)	−1.376	0.196	0.397
NEFA (mEq/L)	0.33 (0.14)	0.39 (0.13)	0.40 (0.14)	0.31 (0.19)	1.413	0.185	0.408
Glucose (mg/dL)	95.3 (7.4)	91.6 (11.6)	92.1 (5.3)	97.0 (12.7)	−1.733	0.111	0.500
Insulin (ulU/ml)	8.1 (3.0)	7.1 (3.2)	6.8 (2.3)	7.8 (3.5)	−1.926	0.080	0.556
HOMA-IR	1.92 (0.72)	1.66 (0.97)	1.54 (0.49)	1.84 (0.80)	−2.252	0.046	0.650

Note: p=p-value from paired t-tests; d=Cohen’s d;

^*Analyses are adjusted for weight change.

A fasting blood draw was taken at 0800h after a minimum 9h fasting period.

DXA – dual-energy x-ray absorptiometry; dHDL – direct high-density lipoprotein; LDL – low-density lipoprotein; NEFA – non-esterified fatty acid; HOMA-IR – Homeostatic Model Assessment of Insulin Resistance

Figure 1. Violin plots showing the distribution and density of changes for the daytime and delayed eating conditions.

Violin plots show the following: (A) weight (kg), (B) resting energy expenditure (REE), (C) respiratory quotient (RQ), (D) trunk-to-leg fat ratio (measured by DXA), (E) total cholesterol, (F) triglycerides, (G) glucose, (H) insulin, and (I) HOMA-IR. For (E-I), a fasting blood draw was taken at 0800h after a minimum 9h fasting period. Daytime: blue, Delayed: red; Black diamonds: individual participant data points (n=12); yellow dots: median.

Regarding fasting biomarkers of metabolic health, total cholesterol and triglycerides increased on the delayed schedule with small and medium effect sizes, respectively (Table 2, Figure 1E–1F). High-density lipoprotein (dHDL) cholesterol and low-density lipoprotein (LDL) cholesterol improved on the delayed schedule, with medium effect sizes, respectively; the dHDL cholesterol difference was significant (Table 2). There was a small increase in adiponectin on the delayed schedule, suggesting slight improvement; non-esterified fatty acid (NEFA) had a small increase on the daytime schedule and a decrease on the delayed schedule (Table 2). Glucose and insulin decreased during daytime eating and increased during delayed eating, both with medium effect sizes (Table 2, Figure 1G–1H). Insulin resistance, measured by Homeostatic Model Assessment of Insulin Resistance (HOMA-IR), also decreased on the daytime schedule and increased on the delayed schedule; the HOMA-IR difference was significant with a medium effect size (Table 2, Figure 1I). Baseline measures before each condition did not differ significantly except for triglycerides, which were slightly higher in the pre-delayed than pre-daytime condition (Table 2, p=0.04).

Circadian Cosinor Analyses

Melatonin and cortisol, central clock markers, showed no significant differences in changes (for all measures in this section, daytime change values are presented first and delayed change values are presented second) in acrophase (melatonin: −0.12±1.60h vs −0.30±7.45h, p=0.94, d=0.02; cortisol: −1.14±6.68h vs −0.72±8.70h, p=0.88, d=0.05); amplitude (melatonin: −22.69±61.16 pg/ml vs 16.88±59.45 pg/ml, p=0.28, d=0.33; cortisol: 1.65±3.40 mg/dl vs 2.27±2.17 mg/dl, p=0.64, d=0.14); or mesor (melatonin: −15.27±37.63 pg/ml vs 10.15±29.58 pg/ml, p=0.21, d=0.38; cortisol: 0.23±4.29 mg/dl vs −0.31±4.89 mg/dl, p=0.76, d=0.09). The metabolic markers leptin, ghrelin, and glucose also showed no differences in changes in acrophase (leptin: 2.02±10.65h vs −3.95±7.42h, p=0.10, d=0.51; ghrelin: −4.93±6.52h vs 2.33±12.39h, p=0.12, d=0.48; glucose: −2.16±10.49h vs −6.78±12.03h, p=0.37, d=0.27); amplitude (leptin: 0.13±2.64 mg/dl vs 2.15±4.74 mg/dl, p=0.25, d=0.35; ghrelin: 19.35±90.94 pg/ml vs 7.73±157.59 pg/ml, p=0.83, d=0.07; glucose: 3.74±11.27 mg/dl vs −3.72±13.05 mg/dl, p=0.14, d=0.46); or mesor (leptin: −2.32±13.58 mg/dl vs 1.93±3.54 mg/dl, p=0.34, d=0.29; ghrelin: 29.57±86.65 pg/ml vs −107.76±321.88 pg/ml, p=0.21, d=0.39; glucose: −0.48±8.34 mg/dl vs 0.49±10.15 mg/dl, p=0.83, d=0.06).

Time Series Analyses

Melatonin did not show suppression in dim light conditions in either eating condition (Figure 2A). Comparisons of the change in mean values for melatonin, cortisol, ghrelin, leptin, and glucose were conducted at each 4-h timepoint (Figure 2A–Figure 2E). Cortisol increased at 0400h on the daytime compared to the delayed condition (3.00±6.94 mg/dl vs −2.45±5.07 mg/dl, p=0.03, d=0.72). No other comparisons were significant.

Figure 2. Mean values of blood hormones and metabolites taken every 4h from 0800h-0400h before and after the daytime and delayed eating conditions.

Plots show the following: (A) melatonin, (B) cortisol, (C) ghrelin, (D) leptin, and (E) glucose. For each eating schedule, data were adjusted by the daily mean. Vertical bars are standard errors.

Discussion

In this carefully controlled study of healthy participants with a BMI of 19 to 27 kg/m², eating on a daytime as compared to a delayed schedule produced improvements in body weight, insulin sensitivity (HOMA-IR), fat oxidation (RQ), trunk-to-leg fat ratio, and fasting glucose, insulin, triglycerides and total cholesterol, while dHDL and LDL cholesterol, adiponectin and REE improved on the delayed condition, all based on effect sizes ranging from d=0.397-1.019. Notably, the circadian phase and amplitude of melatonin, cortisol, ghrelin, leptin, and glucose were not affected by the eating schedules. These findings suggest that overall, delayed eating has adverse effects on body weight and metabolic parameters independent of caloric intake, exercise or sleep; thus, adhering to a sustained daytime eating schedule shows efficacy and feasibility as a behavioral modification for promoting beneficial metabolic and weight management under real world conditions.

We found a significant reduction in HOMA-IR on the daytime compared to the delayed eating condition, suggesting increased insulin sensitivity when eating earlier. Conversely, adiponectin improved slightly with delayed eating, but to a much smaller degree than the improvement in HOMA-IR during the daytime condition. Insulin sensitivity has rarely been measured in timed eating studies in healthy populations: in the first study that measured it, timed eating failed to produce changes in insulin sensitivity [15]. The longer duration of our study compared to the previous study [15] provided more time for an effect to emerge in our healthy population, suggesting the impact of early eating may be gradual and contribute a moderate benefit over the long term by lowering glucose and insulin levels and improving insulin sensitivity [18]. Further, a cross-sectional study using a 5-h oral glucose tolerance test showed that the time by which participants typically consumed 25% of their daily calories predicted improved insulin sensitivity with the Matsuda Index and the QUICKI [19].

The reduction in the trunk-to-leg fat ratio on the daytime compared to the delayed schedule indicates a differential effect of the timing of eating on fat distribution. A higher trunk-to-leg fat ratio was associated positively with triglycerides, total cholesterol, systolic blood pressure, and C-reactive protein, and negatively with dHDL cholesterol in adolescents in the National Health and Examination Survey cohort [20], and also was positively related to diabetes risk in the Women’s Health Initiative sample [21]. Thus, eating earlier may protect against diabetes and cardiometabolic dysfunction.

RQ decreased in the daytime and increased in the delayed condition. High RQ values are indicative of low fat oxidation and high carbohydrate oxidation; thus, an increase in RQ after an overnight fast reflects a poorer ability to oxidize fatty acids and is an important contributor to weight gain and the development of metabolic syndrome and type 2 diabetes [22,23]. The decrease in RQ observed during daytime eating is consistent with a study implementing an early time restricted eating schedule [24] and may be related to our concurrent decreases in insulin. REE decreased during the daytime schedule, in contrast to the various other cardiometabolic improvements observed in this condition, likely due to weight loss, as caloric intake was similar in both conditions. In their early time restricted eating study, Ravussin and colleagues [24] measured REE in a chamber, noting variability across the day, which resulted in no differences in total daily REE; this suggests the effects of timed eating on REE are nuanced and related to differential influences of the thermal effect of food post-prandially throughout the day.

The weight increase in the delayed eating condition was modest but of significance. We note that the weight gain occurred independent of caloric intake and exercise. Because we did not directly measure all components of energy expenditure, we cannot fully account for the potential imbalance between energy intake and energy output, which could have led to weight gain in the delayed condition. Additionally, it is likely that DXA detected small changes in % total body fat but may have been less sensitive in detecting small changes in % lean mass, which could have impacted weight gain in the delayed condition.

The timing of eating did not alter the circadian phase or amplitude of central clock [suprachiasmatic nucleus, SCN] markers (plasma melatonin and cortisol), in agreement with an entrained sleep-wake cycle, and with an experimental timed eating study conducted under a constant routine [17]. Thus, our results, coupled with those of Wehrens et al. [17] indicate the observed metabolic changes occurred independently of the SCN in both inpatient and outpatient settings. We also did not detect differences in the circadian phase or amplitude of plasma leptin, ghrelin, or glucose. By contrast, Wehrens et al. [17] observed a phase delay in glucose, but not in insulin or triglycerides, of nearly 5.7h with a 5-h delay in three isocaloric daily meals under a constant routine. The delayed glucose rhythm in that study did not coincide with changes in insulin levels; thus, it is unclear how this result is related to insulin sensitivity or insulin secretion. Because a 4-h delay in eating on the delayed condition did not appreciably affect the timing or amplitude of peripheral rhythms, the weight and metabolic differences reported above are not explained by circadian changes in blood markers.

We examined only the timing of eating while holding the total number of hours of energy consumption constant at 11 h/day, which allows completion of both eating conditions within waking hours. Compliance was excellent, indicating feasibility of our long duration paradigm for outpatient use. Most adults typically eat across 14-15h/day [25], but our 11-h eating duration was realistic and sufficient for both eating schedules. Wilkinson et al. [26] showed that in persons with metabolic syndrome who typically eat at least 14 h/day, reducing eating duration to 10 h/day in a non-controlled 12-week intervention decreased caloric intake, and produced a 3.3 kg weight loss and improvements in total and LDL cholesterol, adiposity (via bioelectrical impedance), and systolic and diastolic blood pressure, but not in glucose, insulin, dHDL cholesterol, or triglycerides. Notably, timely new evidence also shows a strong positive association between both a later dinner time and a shorter interval of time between dinner and sunset and increased risk of mortality due to COVID-19 infection, possibly through increased inflammation from worsened cardiometabolic health [27]. By contrast, studies investigating Ramadan, which reduces duration and timing by restricting eating to before dawn and at night after sunset, have shown mixed effects on weight and cardiometabolic outcomes. There is some improvement in total cholesterol and triglycerides, and in increased fat oxidation, but inconsistent results for weight loss and improved insulin sensitivity [28]. Thus, during Ramadan, dividing eating duration and timing into early and late periods tempers some of the weight and cardiometabolic benefits observed in our study.

Combining an early schedule with a 6-h restricted eating window (eating from 0800h to 1400h) decreased mean ghrelin levels, improved 24-h glucose levels (measured via continuous glucose monitoring), reduced variability in self-reported hunger ratings, and tended to increase satiety and decrease the desire to eat over a 4-day period, as compared to a 12-h eating condition among adults with overweight and obesity [24,29]. In males with pre-diabetes, this same 6-h schedule produced improvements in insulin sensitivity (measured by an oral glucose tolerance test), β cell responsiveness, systolic and diastolic blood pressure, oxidative stress (8-isoprostane), and appetite [18]. While these results are interesting, eating only 6 h/day and stopping eating by 1400h daily are not feasible for most people to maintain long-term. Because our study and other evidence suggest that maintaining an early eating schedule and a 12-h or more fasting period per day benefits weight management, glucose regulation, and metabolic health, behavioral strategies that limit nighttime eating should be developed for widespread implementation.

Strengths of this study include the highly controlled conditions, elimination of other sources of variance, a realistic eating duration, and high compliance. Participants received all their food to control for caloric and macronutrient intake and were prompted daily to report any compliance issues. Compliance with the schedules and with consumption of the food provided was above 90% for both measures. Sleep-wake times were confirmed with actigraphy and showed that sleep duration, variability, and efficiency were normal and comparable between conditions, and importantly, participants were not sleep deprived. Sleep onset and offset were also close to the prescribed window for both conditions, with the delayed condition producing slightly later bedtimes. Thus, the benefits of the daytime condition can be attributed to the timing of eating and not to other factors.

The changes in body weight and several metabolic markers were medium to large in magnitude. However, the participants nearly all had normal weight and were metabolically healthy, with baseline measures in the normal range. Additionally, because we observed positive weight and metabolic changes across 8 weeks, it seems likely that adhering to daytime eating for a longer duration may result in greater benefits. The study sample size was small as this was the first test of our prolonged duration experimental paradigm. Power was optimized by the within-participants crossover design, so we relied on effect sizes to evaluate our outcomes due to this limitation. Finally, although we carefully controlled for intake by providing meals and by querying participants daily, compliance checks were self-reported and could not be independently verified, and thus this is a study limitation.

This study demonstrates that eating earlier in the day improves weight and several key metabolic outcomes in healthy individuals. Future studies should examine the long-term effects in larger populations with obesity or metabolic syndrome and probe underlying mechanisms. Identifying the metabolic consequences of late eating will provide important insights for the pathophysiology and treatment of obesity, diabetes, and related diseases.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Kelly C. Allison, kca@pennmedicine.upenn.edu.

Materials Availability

This study did not generate new unique reagents.

Data and Code Availability

Original/source data used for the analyses, figures and tables reported in the manuscript will be made available to qualified researchers, with minimal restrictions and in a timely manner.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Participants

Participants were recruited through a variety of advertisements. Adults of all races and ethnicities, ages 21-45 years, with a BMI of 19-27 kg/m², and stable weight (≤4.5 kg change) over the previous 6 months were eligible. Exclusion criteria included: unstable, serious medical conditions; medications linked to weight gain or loss; oral steroid use; diabetes mellitus, cancer, or autoimmune disease; use of recreational drugs, melatonin, diuretics or hypnotics; current weight reduction program; a sleep disorder (determined by questionnaires and actigraphy); working night shifts; habitual waking after 0900h; bedtime later than 2400h; or failure to maintain a regular bedtime and wake time determined by actigraphy. Psychiatric exclusions, determined by the MINI [30], were severe major depressive disorder, and/or suicidal risk; bipolar disorder; current or past psychosis; lifetime eating disorder; or any other severe psychiatric disorder judged to interfere with long-term adherence to the proposed study. Participants were required to maintain their usual exercise levels throughout the study. The final sample size was 12 participants (5 females), with a mean±SD age of 26.3±3.4y. The CONSORT flow diagram is presented in Figure S2. Further details regarding participant characteristics and demographics are reported in the Results section.

Screening Visits

After passing an initial phone screen, participants presented for a screening visit. The Institutional Review Board of the University of Pennsylvania approved this study and all participants provided informed consent. To screen for inclusion criteria, study staff completed a clinical interview assessing typical eating and sleep patterns [Night Eating Syndrome History and Inventory, modified [31], weight history, and psychiatric status (MINI; [30]). Weight (using a calibrated Tanita WB800 digital scale), height (using a Harpenden wall-mounted stadiometer), and waist circumference were measured. Circadian preference was measured with the Composite Scale of Morningness and Eveningness [32]. Sleep patterns and potential disorders were further assessed with the Pittsburgh Sleep Quality Index (PSQI; [33]), the Epworth Sleepiness Scale (ESS; [34]), and the Multivariable Apnea Risk Index (MAP index; [35]) and mood was assessed with the Patient Health Questionnaire–9 items (PHQ-9; [36]). Participants then received logs to record food, exercise bouts, and sleep, and a wrist actigraph with light sensor (Actiwatch Spectrum, Philips Respironics) to wear on the non-dominant wrist for the next 10 days during the screening phase.

The participants were not extreme morning or evening chronotypes (Morningness and Eveningness range, 33-50, 39.08±5.00), and they did not report sleep disorders (PSQI global score=2.75±1.29; ESS score=5.33±3.37; the MAP Index score=0.07±0.06). Mood was in the normal range (PHQ-9 score=0.33±0.78). During the screening period, participants reported (mean±SD) consuming 4.67±1.03 meals or snacks/day and 2081.91±400.95 kcals/day, with their first meal at 09:13±1:06h, their last meal at 19:59h±1:10h, and an eating duration of 10:30±1:31h/day. Mean±SD actigraphic sleep variables during screening were: sleep duration (7.02±1.05h), sleep onset (23:44±0.97h), sleep offset (8:10±0.97h), sleep onset latency (21.14±17.69min), sleep efficiency (83.81±5.06%), and wake after sleep onset (43.53±15.25min).

Actigraphs were also then continuously worn for the 18 weeks of the study to determine sleep onset, offset, duration and efficiency. Actigraphy was also used to determine amounts of sedentary, light, moderate and vigorous activity, confirm periods of deliberate physical activity reported in participants’ logs [37], and determine light levels as detected at the wrist.

Once the regularity of their daily routines and their ability to comply with procedures was confirmed, participants underwent a medical history and physical. If they passed this medical screen, participants completed their baseline inpatient assessment and, afterwards, were randomized to the daytime or delayed eating schedule for their first 8-week eating condition. At the end of the 8 weeks, participants returned for their second inpatient assessment, followed by a 2-week washout during which they ate their own food on their typical schedule. They then completed a third inpatient assessment, which served as baseline for their second 8-week eating condition—either daytime or delayed, based on their randomized order—followed by the fourth and final inpatient assessment (Figure S3).

METHOD DETAILS

Inpatient Assessment Protocol

Each assessment included a 28-h stay at the Center for Human Phenomic Science (CHPS) at the Hospital of the University of Pennsylvania. The night prior to these inpatient assessments, participants fasted from 2300h until the 0800h blood draw for each condition’s baseline assessment (visits 1 and 3), and after the delayed eating condition, producing a 9-h fast, and from 1900h until the blood draw at 0800h following the daytime eating condition, producing a 13-h fast. We chose to have participants adhere to their assigned eating schedule given that the duration of the fasting period was long (≥9h) in both conditions. Previous studies found the following: 1) no significant differences in glucose or insulin levels between 7h to up to 13h of fasting [38,39], 2) fasting glucose did not differ for <8h versus ≥8h of fasting [38], and 3) glucose was stable after only 3h of fasting [40]. As such, the fasting period differences in our study conditions would not appreciably impact our fasting values for insulin, glucose, or HOMA-IR.

Participants arrived at 0700h, anthropometrics were measured, and drug and pregnancy tests (for females) were performed. They then were housed in a <20 lux room (to avoid suppression of melatonin secretion; [41]) and placed in a supine position. An indwelling intravenous line was initiated at 0730h. Blood was drawn at 4-h intervals [0800h (fasting), 1200h, 1600h, 2000h, 2400h, and 0400h] to measure amplitude and phase of circadian rhythms of glucose, leptin, ghrelin, melatonin and cortisol [42–44]. In addition, fasting levels (0800h) of insulin, adiponectin, cholesterol, triglycerides, and NEFAs were measured on morning 1. HOMA-IR was calculated using the 0800h sample and the following equation: fasting serum insulin (microU/mL) × fasting blood glucose (mg/dL)/405 [45].

On the second morning following the blood draws and overnight fast (fasting began by 2130h before the start of each eating condition and after the delayed condition, and 1900h after the daytime condition), indirect calorimetry was performed using a metabolic cart at 0800h to assess REE and fuel oxidation (RQ) (Parvo Medics TrueOne 2400). Before each measurement, the metabolic cart was calibrated with reference gas and with a flow meter calibration. After achieving a steady state (15 minutes), expired gases were collected for 30 minutes and used to calculate metabolic rate. We and other researchers have utilized this metabolic cart in prior studies—it shows validity and reliability for measurement of REE and RQ [46–48]. Finally, body composition was measured using DXA (Hologic Discovery Wi Bone Densitometer).

Eating Conditions

Prescribed meals and snacks were provided twice per week during each 8-week eating condition by participant pick-up or delivery by staff. Personalized menus accompanied the three meals and two snacks assigned for each day. Participants noted each food item consumed, the time and any modifications. The CHPS metabolic kitchen staff provided portion size training at baseline. Participants were permitted to drink non-caloric beverages (e.g., water, diet beverages, black coffee) outside of their assigned eating windows and noted all beverages consumed on their logs. For both conditions, the diet consisted of approximately 55% carbohydrate, 15% protein, and 30% fat. Foods were chosen to match participants’ preferences to maximize consumption of the provided food. Energy needs were calculated for each participant using the Harris-Benedict equation [49]. Staff also provided a new actigraph and collected the food logs during these food provisions to monitor compliance continually.

If participants were unable to consume a meal or snack as provided, they sent a picture of the meal or snack they consumed as a substitute using an electronic device. Uneaten food was returned to the metabolic kitchen, when possible. Kitchen staff pre-weighed all provided items, post-weighed any returned items, and used participant logs to enter food/drink intake into the Nutrition Data System for Research (NDSR) and compute each participant’s daily caloric and macronutrient intake.

Compliance

Frequent contact between study staff and participants allowed for feedback about the acceptability of the food items and difficulties with adherence to the eating and sleep-wake schedule (initiating sleep between 2200-2400h and waking by 0930h). To further monitor compliance, participants received the following six queries every morning via the RedCAP data capture system: 1. What time did you go to bed last night? 2. What time did you wake up this morning? 3. How many meals and snacks did you consume? 4. Did you eat any food outside of what was provided? (If yes, please provide details/send picture.) 5. Did you eat outside of the time window (0800h-1900h for daytime condition and 1200h-2300h for delayed condition)? 6. How many minutes did you exercise yesterday?

Assays

Blood was collected in pre-cooled vacutainer tubes containing EDTA and kept on ice until centrifugation at 4°C and storage at −80°C. Plasma hormones were measured by radioimmunoassay in duplicate. The precision of assays was as follows: cortisol (MP Biomedical, Solon, OH) intra-assay coefficient of variation (CV) 3.78%, inter-assay CV 11.3%; and melatonin (Tecan, Baldwin Park, CA) intra-assay CV 4.68%, inter-assay CV 26.3%. Leptin, ghrelin, adiponectin, and insulin (EMD Millipore, Billerica, MA) had CVs as follows: leptin intra-assay 3.1%, inter-assay 13.5%; total ghrelin intra-assay 4.49%, inter-assay 13.9%; adiponectin intra-assay 8.0%, inter-assay 10.4%; and insulin intra-assay 4.99%, inter-assay 11.3%. All assays had CVs within the acceptable range. Glucose was analyzed via a YSI 2900 glucometer, and NEFA were measured spectrophotometrically using Wako Life Sciences reagents (Richmond, VA). Triglycerides, total cholesterol, dHDL cholesterol and LDL cholesterol were measured using a Roche Cobas c311 Automated Clinical Chemistry Analyzer (Indianapolis, IN).

QUANTIFICATION AND STATISTICAL ANALYSIS

Descriptive statistics characterized the sample and outcome measures, including mean, standard deviation, and standard error of the mean, as indicated in the methods, results, tables, and figures. Participants were randomized based on a code generated by a random number generator. The final sample size for analyses was n=12 participants (as explained in the results, one participant’s data were excluded). Because this was the first study using this specific prolonged, rigorous protocol, Cohen’s d effect sizes were used for the primary outcomes, using the following ranges: ≥0.2 small, ≥0.5 medium, ≥0.8 large [50]. The Shapiro-Wilk test did not reject variable normality. Paired t-tests were used to compare delta values (before vs after each eating schedule) for anthropometric, compliance, and metabolic measures between the daytime and delayed schedules, with p<0.05 considered statistically significant (using two-tailed tests); for fasting hormones and metabolites, paired t-tests after adjusting for weight change were used. Cosinor analysis compared the effect of timed eating schedules on diurnal rhythm phase markers of the central clock, (melatonin and cortisol), biomarkers of appetite and metabolism (leptin, ghrelin, and glucose) and on actigraphic activity counts in 10-minute intervals (Figure S1). We used a cosinor model with a fixed 24-h period, to derive amplitude (half the peak-to-trough difference), acrophase (peak time), and mesor (mean) and compared changes in these values with paired t-tests (using two-tailed tests). Time series data were evaluated with independent t-tests for change values at each timepoint, with p<0.05 considered statistically significant (using two-tailed tests). Given this was an initial study, we did not correct for multiple comparisons in the above analyses. All statistical analyses and plotting were performed within the R statistical environment (version 3.5.1).

ADDITIONAL RESOURCES

The ClinicalTrials.gov Identifier for this trial is: NCT04414644.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Bacterial and Virus Strains
Biological Samples
Serial human blood samples	This paper	N/A
Chemicals, Peptides, and Recombinant Proteins
Critical Commercial Assays
Cortisol	MP Biomedical	Cat# 07-221102
Melatonin	TECAN US	Cat# RE29301
Leptin	EMD Millipore	Cat# HL-81K
Ghrelin	EMD Millipore	Cat# GHRT-89HK
Adiponectin	EMD Millipore	Cat# HADP-61HK
Insulin	EMD Millipore	Cat# HI-14K
Glucose	YSI 2900 glucometer
NEFA	Wako Life Sciences	Cat# 999-34691 Cat# 995-34791 Cat# 991-34891 Cat# 993-35191
Triglycerides	Roche Cobas c311 Automated Clinical Chemistry Analyzer	Cat# 20767107322
Total cholesterol	Roche Cobas c311 Automated Clinical Chemistry Analyzer	Cat# 03039773190
dHDL cholesterol	Roche Cobas c311 Automated Clinical Chemistry Analyzer	Cat# 07528566190
LDL cholesterol	Roche Cobas c311 Automated Clinical Chemistry Analyzer	Calculated by software from: Cat# 20767107322 Cat# 03039773190 Cat# 07528566190
Deposited Data
Experimental Models: Cell Lines
Experimental Models: Organisms/Strains
Human subjects	This paper	N/A
Oligonucleotides
Recombinant DNA
Software and Algorithms
Other
Resting Energy Expenditure and Respiratory Quotient	Parvo Medics TrueOne 2400	N/A
Dual-energy X-ray absorptiometry	Hologic Discovery Wi Bone Densitometer	N/A

Highlights.

• An early (0800h-2100h) versus delayed (1200h-2300h) eating schedule reduces weight

• Eating early vs later improves insulin resistance, and insulin and glucose levels

• Eating early vs later improves fat oxidation, cholesterol and trunk-to-leg fat ratio

• An early eating schedule benefits weight and cardiometabolic health