Preliminary Evaluation of Dim Light Melatonin Onset as an Indicator of Nighttime Eating in Adolescents

Introduction

Circadian rhythms, controlled by the suprachiasmatic nucleus, influence the timing of wake-sleep, hormone secretion, and eating behaviors within a 24-hour cycle in humans. These circadian rhythms facilitate diurnal (vs. nocturnal) behaviors, including increased alertness, activity, and eating during daylight, and sleep during night. Thus, circadian rhythms help coordinate the timing of biological night. Biological night refers to the internal biological changes in body temperature and hormones (e.g., melatonin, cortisol) that prepare the body for sleep. Chronotype refers to individual differences in the typical or preferred timing of sleep and activity during a 24-hour period. People with late sleep chronotypes prefer later sleep-wake timing (i.e., later preferred time in bed potentially accompanied by later preferred wake time) compared to people with early sleep chronotypes. Several biological and environmental factors contribute to individual differences in sleep chronotypes (e.g., societal impact of late evening activities and cell phone use), including age. Adolescence is marked by an increased preference for a late sleep chronotype¹ and is a high-risk period for the development of maladaptive eating behaviors and obesity. A late sleep chronotype in adolescence has been linked to late meal timing, excess energy intake, poor diet quality, and skipping breakfast^2–5, and these maladaptive eating behaviors may help to explain elevations in body mass index (BMI) among adolescents with a late sleep chronotype^6–8.

Late meal timing is one possible mechanism explaining the relationship between late sleep chronotype and increased risk for overweight and obesity. In adults, late eating led to increased hunger, increased ghrelin:leptin ratios, decreased energy expenditure upon awakening, changes in lipid metabolism⁹, and decreased glucose tolerance¹⁰. The co-occurrence of food intake and high levels of melatonin, a likely consequence of late eating, is linked to adverse metabolic effects (e.g., glucose intolerance)^10,11,12. Furthermore, later meal timing also disrupts sleep duration and sleep quality¹³ contributing to additional negative metabolic effects¹⁴. Despite initial evidence that melatonin onset likely plays a pivotal role in understanding whether later meal timing confers risk for overweight/obesity, current definitions of nighttime eating do not account for individual differences in circadian rhythms/chronotype. Nighttime eating behaviors are typically defined in research studies as consuming ≥ 25% of daily calories after the evening meal or after a predetermined time of day (e.g., 20:00, 21:00, 22:00)^15–17. These operational definitions of nighttime eating are consistent with the diagnostic criteria for Night Eating Syndrome, although they may not capture non-diagnostic but nevertheless maladaptive behaviors that confer risk for poor health outcomes outside of a formal clinical diagnosis of Night Eating Syndrome¹⁸. When not addressed, maladaptive eating in adolescence could persist or worsen in adulthood increasing later risk for overweight/obesity and poor health outcomes. Thus, it is critical to identify and intervene on nighttime eating behaviors that increase risk for the development of metabolic disorders.

Given that deleterious metabolic effects occur when food intake co-occurs with high melatonin levels¹⁰, we propose a novel approach that considers nighttime eating behaviors as eating after the onset of biological night. Melatonin secretion occurs during the dark phase of the 24-hour day and thus signals biological night¹⁹. Recent articles have advised against using self-reports of chronotype as proxy measures for biological circadian time phase due to inconsistent evidence of convergent validity between self-report and biological measures^20,21, particularly in high-school age adolescents²². Dim light melatonin onset (DLMO) identifies the time of melatonin secretion using saliva samples. DLMO has previously been used in adult samples to more objectively operationalize meal timing relative to the start of biological night. Later meal timing has been defined as interval (in hours) between caloric midpoint (i.e., time of day when ≥ 50% of calories are consumed) and the start of biological night as assessed by DLMO¹¹. An alternative definition is interval (in hours) between the last eating occasion and DLMO¹². These approaches have conjured some support in adults as individuals with a higher body fat percentage ate most of their calories about an hour closer to the time of melatonin onset compared to lean individuals¹¹. Similarly, a shorter interval between the last meal and melatonin onset was associated with consuming more meals and a greater percentage of calories from carbohydrates and sugar¹². These definitions of meal timing relative to DLMO have not been investigated in adolescents, and no prior research has attempted to define nighttime eating as eating after the onset of melatonin.

The purpose of this exploratory, proof-of-concept study was to evaluate the utility of defining nighttime eating using a biological assessment of circadian phase, DLMO, and examining if individuals who ate before versus after DLMO differed in their BMI, waist circumference, physical activity, and eating behaviors. This exploratory study defined nighttime eating based on timing of last eating episode relative to an objective circadian marker of the start of biological night, DLMO. First, we evaluated the convergent validity of the proposed definition of nighttime eating by comparing individuals who ate after (vs. before) DLMO on two prior definitions of late meal timing: 1) interval between caloric midpoint and DLMO, and 2) interval of last eating episode and DLMO. We expected that individuals with nighttime eating (i.e., eating after DLMO) would have shorter intervals between melatonin onset and their caloric midpoint and last eating episode compared to individuals without nighttime eating. Second, we aimed to test our hypotheses that individuals who ate after DLMO would report increased dietary intake, decreased physical activity, and an elevated BMI percentile relative to individuals who ate before DLMO. Sleep indices were included as secondary outcomes with no specific hypotheses.

Methods

The current study was a secondary data analysis of a larger, ongoing parent project focused on sleep/circadian-related risk for substance use in adolescents (R01 DA044143). The parent study included an observational phase and an experimental manipulation of sleep timing. Only data from the observational phase were included in the current study.

1.1. Participants

Adolescents aged 16–18y who were enrolled in 11th or 12th grade were eligible to participate in the parent study (N=136). Adolescents were recruited from the community via paid social media advertisements, email listservs, study flyers placed in schools, and word of mouth/referrals. Adolescents were not eligible if they were: diagnosed with significant or untreated medical conditions that interfere with sleep or well-being (e.g., renal failure, liver disease); taking medications other than selective serotonin reuptake inhibitors/serotonin and norepinephrine reuptake inhibitors (SSRIs/SNRIs) known to interfere with sleep or reward sensitivity (e.g., benzodiazepines, antipsychotics, beta-blockers); endorsing a lifetime history of bipolar, psychotic, or substance use disorders (other than alcohol or cannabis use disorder); engaging in daily nicotine use or recreational drug use (other than cannabis) in the past month; diagnosed with a current sleep disorder (other than delayed sleep disorder or insomnia); experiencing/reporting any conditions that would interfere with MRI protocols (e.g., implanted medical devices, braces). To be included in the current study, adolescents needed to have a valid measurement of DLMO and eating behavior reported via dietary recall aligned to the prior 24-hour period (N=68).

1.2. Procedures

Adolescents completed a one-week observational ecological momentary assessment (EMA) protocol, which included 24-hour wrist-worn accelerometry, six signal-contingent surveys, and event-contingent surveys upon waking up and prior to going to sleep. All data were collected during the school year when participants were subject to school start times. Following the EMA protocol, adolescents participated in a weekday (Wednesday or Thursday) overnight lab visit to assess DLMO. While in the lab, participants completed a dietary recall for the 00:00–23:59 24-hour period of the previous day. Thus, participants reported on eating that occurred in their natural environment on the day before DLMO assessment. All study procedures were approved by the University of Pittsburgh Institutional Review Board. See Figure 2 for a timeline of study procedures.

Figure 2. Timeline of Study Assessments.

Note. EMA=ecological momentary assessment; DLMO= dim-light melatonin onset. During a seven-day observational phase, adolescents completed wrist-worn actigraphy to assess daily physical activity and sleep. Following the seven-day observation period, on day eight, adolescents attended a lab visit. Actigraphy data was collected on the eight day, although those data were not used in the current study. During this lab visit, adolescents completed a dietary recall for the prior day (Day 7 00:01–23:59) and provided saliva samples to calculate the time of melatonin onset (i.e., DLMO).

1.3. Measures

1.3.1. Dim Light Melatonin Onset (DLMO)

Circadian phase was assessed using salivary DLMO²³; saliva samples were collected in Salivettes under dim light conditions (<10 lux at every angle of gaze confirmed with a light meter) during the overnight lab visit. Saliva was collected in half-hour increments in the six hours leading up to and one hour following each participant’s habitual bedtime. Standard DLMO sampling procedures were followed (e.g., remaining seated, not consuming food/drink within ten minutes of sampling, rinsing with water following food or drink intake, and avoiding consumption of caffeine, bananas, or chocolate). Saliva samples were assayed by Solid Phase (Portland, ME) using radioimmunoassay kits for salivary melatonin, and DLMO was calculated as the clock time when salivary levels of melatonin exceeded 4 pg/m1²⁴. In young adult data, test-retest reliability of DLMO is strong with intraclass correlation coefficients ranging from 0.85 for one month test-retest reliability to 0.91 for one night test-retest reliability^28,37.

1.3.2. Physical Activity

Wrist-worn accelerometry data were collected via an Actiwatch Spectrum Plus (Respironics/Philips, Bend, OR). Adolescents wore the Actiwatch on their non-dominant wrist for seven days during the observational EMA period. Activity count data were extracted and processed via an R script that created variables to indicate whether the count data in each 60-s epoch corresponded to moderate-to-vigorous physical activity (MVPA) using a cut point of 418²⁵. Only data collected during waking hours were extracted to examine day-time physical activity, and days with <10 hours of valid wear time were excluded. Activity data were summarized as the average number of minutes per day spent in MVPA, as well as time spent in MVPA in the morning (06:00–11:59), afternoon (12:00–16:59), and evening (17:00–22:59).

1.3.3. Sleep

Data from the Actiwatch were used to estimate sleep indices, including total sleep time, sleep efficiency (i.e., ratio of total sleep time to time in bed), sleep midpoint (i.e., midpoint between sleep and wake onset), and social jet lag (i.e., individual differences between weekday and weekend sleep midpoints) that occurred, on average, during the 7-day EMA observational phase. Prior to scoring via Actiware, data were cleaned using a standardized approach similar to Patel et al. (2015)²⁶, assigning a rest interval based on the concordance between available inputs (event marker, white light intensity, sleep diary, and activity). Sleep data from the overnight lab visit were not included as the goal was to capture sleep behaviors in the natural (i.e., non-lab) environment.

1.3.4. Dietary Recall

Dietary intake in the most recent midnight-to-midnight 24-hour interval prior to DLMO assessment was collected via the Automated Self-Administered 24-hour Dietary Assessment Tool (ASA24) during the weekday lab visit. Within-day nutritional data from food and beverages consumed were extracted from ASA24, including total calories (kcal) consumed in a day, as well as kcal from protein, fat, and carbohydrates. ASA24 data were also used to identify the timing of each participant’s caloric midpoint (i.e., time when at least 50% of daily calories were consumed) and last eating occurrence, which was defined as the final time of day that each participant ate something caloric (i.e., an episode of drinking water or other non-caloric beverage was not considered the last eating episode). Proposed definitions of final eating episodes recommend a specific threshold (e.g., 210kj or 50kcal) to be considered an eating episode²⁷; however, the current study identified the final eating episode as the last instance of eating regardless of calories consumed. The final eating episode, regardless of caloric intake, was utilized in the current study because consumption of any food signals an awake state to the body which is at odds with the body’s preparations for sleep during biological night.

1.3.5. Anthropometrics

Participants’ weight and height were measured in light indoor clothing by a trained assessor. BMI was calculated, and BMI percentiles were estimated using the Centers for Disease Control calculator. Waist circumference was measured via a tape measure placed at belly button height.

1.4. Statistical Analyses

Analyses were completed in R Version 4.3.1. Final eating occurrence was categorized as occurring “Before DLMO” or “After DLMO” indicating whether each participant ate (as reported via ASA24) before or after laboratory-based DLMO. Groups (Before vs. After DLMO) were compared on BMI, BMI%ile, waist circumference, dietary information, sleep indices, average MVPA, and interval (in decimal hours) between DLMO and time of caloric midpoint and time of last eating episode using Welch’s two-sample t-tests and Cohen’s d effect sizes. Given the preliminary nature of this work and the small sample size, effect sizes are considered rather than inferential tests of significance.

Results

Mean ± SD age and BMI percentile were 17.4 ± 0.6 and 67.6 ± 28.5, respectively. Most participants reported being assigned female at birth (65.6%), non-Hispanic (94.6%), and White (81.7%) (Table 1).

Table 1.

Demographics and Clinical Characteristics

Variable	M(SD)	Before DLMO (n=48)	After DLMO (n=20)	Comparison Test Statistic (df)
Age	17.46 (0.62)	17.48 (0.62)	17.43 (0.61)	0.32 (36.31)
BMIa	25.08 (5.10)	25.61 (5.09)	23.88 (5.03)	1.27 (36.96)
BMI %ileb	67.44 (28.15)	71.35 (27.42)	58.18 (28.39)	1.71 (32.88)
DLMO	21.56 (1.35)	21.92 (1.28)	20.70 (1.09)	4.00 (41.49)***
	N (%)
Sex
Female	43 (63.24%)	29 (60.42%)	14 (70.00%)	0.22 (1)
Male	25 (36.76%)	19 (39.58%)	6 (30.00%)
Race
Asian	1 (1.47%)	1 (2.08%)	0 (0.00%)	3.32 (4)
Biracial	3 (4.41%)	2 (4.17%)	1 (0.05%)
Black	8 (11.76%)	6 (12.50%)	2 (10.00%)
White	56 (82.35%)	40 (83.33%)	16 (80.00%)
Ethnicity
Hispanic	3 (4.41%)	3 (6.25%)	0 (0.00%)	0.25 (1)
Not Hispanic	65 (95.59%)	45 (93.75%)	20 (100.00%)
Grade
11th grade	24 (35.29%)	16 (33.33%)	8 (40.00%)
12th grade	44 (64.71%)	32 (66.66%)	12 (60.00%)	0.06 (1)
Weight Categoriesb
Healthy Weight Range	38 (55.88%)	24 (50.00%)	14 (70.00%)	2.48 (2)
Obesity Range	13 (19.12%)	10 (20.83%)	3 (15.00%)
Overweight Range	13 (19.12%)	11 (22.92%)	2 (10.00%)
Underweight Range	0 (0.00%)	0 (0.00%)	0 (0.00%)

Note. BMI= body mass index; BMI %ile= body mass index percentile; DLMO= dim light melatonin onset; Healthy Weight Range= BMI >=5^th percentile and < 85^th percentile; Obesity Range= BMI >= 95^th percentile; Overweight Range= BMI >=85^th percentile and < 95^th percentile; Underweight Range= BMI < 5^th percentile. DLMO values represent decimal, 24-hour clock times. Thus, on average, participants in this study started secreting melatonin around 9:30pm (DLMO=21.56). Welch’s two sample t-tests were used to compare the Before vs. After DLMO groups on continuous outcome variables and chi-square analyses were used for categorical outcome variables.

^a3 participants missing BMI measurement

^b4 participants missing BMI percentile

^*p < .05

^**p <.01

^***p <.001

During the 24 hours prior to DLMO collection, most adolescents’ (n=48; 70.6%) last eating episode occurred before DLMO^a. Medium-to-large effect sizes highlighted that the After DLMO group had shorter intervals between DLMO and the caloric midpoint (Cohen’s d=0.63) and the last eating episode (Cohen’s d=2.33) than the Before DLMO group. Small-to-medium effects were observed, highlighting that the After DLMO group consumed greater total kcal, as well as more kcal (in g) from protein, carbohydrates, and added sugars (Cohen’s d=0.34–0.46) and had lower BMI percentiles (Cohen’s d=0.47) than the Before DLMO group. The After DLMO group engaged in higher levels of MVPA (Cohen’s d=0.68). There were no observed differences in sleep duration, sleep efficiency, or social jet lag between groups, with negligible to small effect sizes (each d < 0.20). The Before DLMO group had a later sleep midpoint compared to the After DLMO group (Cohen’s d=0.60). See Table 2 for additional details.

Table 2.

Comparing Before- and After-DLMO groups

	Before DLMO n = 48		After DLMO n = 20
	M (SD)			T (df)	p-value	Cohen’s d
Alternative DLMO-based Definitions
Caloric Midpoint Interval	−6.24 (2.90)	−3.95 (4.29)		2.20 (26.54)	0.04	0.63
Last Eating Episode Interval	−2.13 (1.64)	0.95 (0.89)		9.93 (61.05)	<0.001	2.33
Anthropometrics
BMIa	25.61 (5.09)		23.88 (5.03)	1.27 (36.96)	0.21	0.34
BMI %ileb	71.35 (27.42)		58.18 (23.39)	1.71 (32.88)	0.10	0.47
Waist (cm)b	76.28 (22.75)		74.65 (17.48)	0.31 (47.16)	0.76	0.08
Total MVPAc (min/day)	188.28 (63.60)		240.15 (86.92)	−2.36 (25.99)	0.03	0.68
Morning MVPAc	51.34 (21.74)		68.27 (28.97)	−2.03 (26.41)	0.04	0.66
Afternoon MVPAc	67.86 (18.38)		87.07 (40.78)	−1.42 (10.78)	0.18	0.61
Evening MVPA	83.20 (40.56)		106.88 (47.68)	−1.19 (8.70)	0.26	0.53
ASA24 Daily Dietary Summary
Total kcal	1710.01 (800.44)		2052.40 (914.73)	−1.46 (31.75)	0.15	0.40
Protein (kcal)	266.24 (161.76)		347.16 (192.16)	−1.65 (30.79)	0.12	0.46
Fat (kcal)	648.45 (343.8)		697.77 (360.45)	−0.52 (34.15)	0.61	0.14
Carbohydrates (kcal)	814.72 (406.36)		1026.48 (505.52)	−1.66 (29.74)	0.12	0.46
Solid Fats (kcal)	312.21 (217.71)		351.00 (273.24)	−0.56 (29.54)	0.58	0.16
Added Sugar (kcal)	50.08 (36.08)		66.68 (59.16)	−1.17 (25.10)	0.25	0.34
Overall Sleep
Midsleepc	4:09 (0.86)		3:41 (0.74)	2.29 (38.21)	0.03	0.60
Efficiency (%)c	84 (0.05)		85 (0.05)	−0.53 (29.92)	0.60	0.20
Duration (hours)c	6.38 (0.57)		6.44 (0.97)	−0.25 (23.17)	0.80	0.08
Social Jet Lagc	1.85 (1.05)		1.92 (1.19)	−0.22 (29.57)	0.82	0.06

Note. BMI= body mass index; BMI %ile= body mass index percentile; Waist= waist circumference; MVPA= moderate-to-vigorous physical activity. Results from Welch’s two-sample t-test are presented. Missing BMI, BMI%ile, Waist circumference, and MVPA were data were coded as missing and could not be imputed. On average, sleep data were available for 4.85 days.

^a3 participants missing BMI measurement

^b4 participants missing BMI percentile

^c1 participant missing MVPA and sleep data

Discussion

The purpose of this exploratory, proof-of-concept study was to test a novel methodology for defining nighttime eating using DLMO, an objective circadian marker of the start of biological night. We aimed to examine whether nighttime eating, defined as eating after DLMO, is associated with maladaptive eating, physical activity, and elevated BMI percentile. Study results should be interpreted cautiously as results are preliminary and require further replication.

Consistent with prior literature linking late meal timing with maladaptive eating behaviors in adolescents, eating after DLMO is likely associated with eating more overall total calories within a day, as well as more calories from protein, added sugar, and carbohydrates^2–5. These results provide initial support that using the proposed definition of nighttime eating could be useful in identifying adolescents who may be more likely to consume high-energy, low nutrient dense foods. In further support of the proposed definition of nighttime eating, there was convergence between the proposed definition of nighttime eating and previous definitions of late meal timing using DLMO known to be associated with increased body fat and poor dietary intake in adults^11,12. While convergence of these measures is important, future research with a sufficiently powered sample should test whether the proposed definition of nighttime eating (i.e., eating after DLMO) provides any incremental validity in concurrently and prospectively predicting outcomes of interest, including BMI, body composition, and dietary intake.

The finding that individuals eating after DLMO was associated with a lower BMI percentile was unexpected, although these results may be accounted for by this group also reporting higher levels of physical activity. It is also possible that one day of dietary recall may not reflect usual dietary patterns and nighttime eating as it is unclear how frequently adolescents ate after DLMO and how the frequency of eating after DLMO infers risk for elevated BMI. It was also unexpected that adolescents with and without nighttime eating did not differ in total sleep duration, sleep efficiency, or social jet lag, on average. This is inconsistent with prior research linking late meal timing with poor sleep behaviors¹³. Adolescents who ate before DLMO reported a later sleep midpoint indicating that they went to bed later and/or woke up later compared to adolescents who ate after DLMO. It is possible that sleep-wake timing was impacted by the timing of structured activities for adolescents (e.g., extracurricular activities before and after school, timing of the school day and bus pick up). For example, participating in sports that involve evening practices and games would likely impact sleep-wake timing, physical activity, mealtimes, and potentially BMI. Future research that attempts to parse out differential impacts of various evening activities that lead to later sleep and eating schedules (e.g., structured sport activities vs. opting to stay up later for optional, non-structured activities) on metabolic outcomes of interest is warranted. Of clinical relevance, the mean total sleep time was approximately 6.5 hours, which is both lower than the recommended 8–10 hours for adolescents³⁹and also consistent with the insufficient sleep reported in other adolescents facing early school start times³⁸.

Several limitations are noteworthy. First, the sample size was relatively small with limited racial/ethnic diversity, which restricted our power to observe statistical significance for the observed effect sizes and limits generalizability. Second, the protocol included only one 24-hour dietary recall. It is possible that individuals vary day-to-day in terms of whether their last eating episode occurred before or after DLMO. A higher proportion of days with nighttime eating to days without nighttime eating may represent increased risk for poor metabolic outcomes and increased BMI as demonstrated in adult samples^9–14. Future research examining dietary information across multiple days relative to DLMO is warranted. Third, dietary recalls are potentially biased by retrospective recall. Although dietary recall assessors were trained to adequately assist with accurate recalls, it is possible dietary recall information is biased. Fourth, although DLMO is thought to be a stable biological indicator of circadian phase^28,37, multiple studies in adults and adolescents have demonstrated shifts from weekdays to weekends^29–32. While the observed weekday DLMO value may not be representative of all days, the current study has a strong weekday estimate of biological night for the period aligning with dietary recalls which is also a weekday. Future research should consider collecting DLMO at multiple times, in addition to multiple samplings of dietary intake, to account for the weekend to weekday variability in DLMO and eating. Finally, substance use is known to impact eating and sleep behaviors^33–35. The current sample of adolescents had very infrequent substance use, although future research in samples with higher substance use should consider controlling for the impacts that alcohol and/or cannabis might have on results given that these substances are most commonly used in adolescents³⁶

Despite limitations, the current study demonstrated several strengths as an exploratory, proof-of-concept approach that could have future research and clinical implications. This study was one of the first to examine the effects of meal timing relative to DLMO. Strong assessment methods were utilized across study variables, including 24-hour dietary recall to assess dietary intake and actigraphy to assess physical activity and sleep in participants’ natural environment. Though DLMO may not be routinely assessed in clinical settings due to intensive sampling procedures, identifying last eating episodes relative to the onset of melatonin secretion (via DLMO) may facilitate research focused on mechanisms linking excess energy intake, poor diet quality, and metabolic risk in adolescents. If reliable norms become available for average timing of melatonin onset across different ages, sexes, and clinical populations, providers could provide general psychoeducation on timing meals to not overlap with the start of biological night. If future technological advancements or other biomarkers reduce the burden of identifying the onset of biological night, personalized treatment approaches could allow providers to instruct patients to time their meals before their personal average time of DLMO onset.

Conclusion

We provided initial evidence that a novel approach to define nighttime eating relative to an objective indicator of biological night (before vs. after DLMO) was associated with different energy intake profiles and corresponded with alternative approaches of operationalizing late meal timing known to be associated with poor metabolic outcomes. Individuals who ate after DLMO were observed to have small-to-moderate effect sizes indicating that they consumed more total calories, as well as calories from protein, carbohydrates, and sugar. If replicated, these results suggest that regular eating after DLMO may be associated with excess caloric ingestion, which over time, if not balanced by increased energy expenditure through physical activity, may lead to increases in relative BMI. Further replication in larger samples of adolescents with additional samplings of DLMO and dietary recall are warranted.

Figure 1. Definitions of nighttime eating.

Note. The current study defines nighttime eating as any eating occurring after the onset of melatonin secretion. During the day, melatonin levels are low and start to increase later in the day. The figure depicts 21:00 as the time of melatonin onset; however, this time is variable across individuals in the current study. Alternative definitions of nighttime eating are listed.