Altered neuronal response to visual food stimuli in adolescents undergoing chronic sleep restriction

Abstract

Study Objectives

Poor sleep in adolescents can increase the risk of obesity, possibly due to changes in dietary patterns. Prior neuroimaging evidence, mostly in adults, suggests that lacking sleep results in increased response to food cues in reward-processing brain regions. Needed is a clarification of the mechanisms by which food reward processing is altered by the kind of chronic sleep restriction (SR) typically experienced by adolescents. This study aimed to elucidate the impact of sleep duration on response to visual food stimuli in healthy adolescents using functional neuroimaging, hypothesizing increased reward processing response after SR compared to a well-rested condition.

Methods

Thirty-nine healthy adolescents, 14–17 years old, completed a 3-week protocol: (1) sleep phase stabilization; (2) SR (~6.5 h nightly); and (3) healthy sleep (HS) duration (~9 h nightly). Participants underwent functional MRI while performing a visual food paradigm. Contrasts of food versus nonfood responses were compared within-subject between conditions of SR and HS.

Results

Under SR, there was a greater response to food stimuli compared to HS in a voxel cluster including the left ventral tegmental area and substantia nigra. No change in food appeal rating due to the sleep manipulation was detected.

Conclusions

Outcomes of this study suggest that SR, as commonly experienced by healthy adolescents, results in the elevated dopaminergic drive of the reward network that may augment motivation to seek food in the context of individual food appeal and inhibitory profiles. Countermeasures that reduce food salience could include promoting consistent HS habits.

Graphical Abstract

Statement of Significance.

Poor eating habits and obesity are increasingly important problems facing adolescents today. Habits developed during this critical developmental stage can carry on to adulthood. There is evidence, mostly in adults, that shortened sleep can influence dietary choices by altering reward processing. This study focused on the changes in neuronal response to visual food stimuli in adolescents due to the sort of chronic sleep restriction they commonly experience. We found that structures responsible for dopaminergic drive of the reward network increased response to viewing food images under sleep restriction. The resulting potential for an increased desire for food might be mitigated by improving sleep habits in adolescents and strategically restricting access to foods when stressors may lead to shortened sleep.

Introduction

Sufficient sleep plays a significant role in maintaining adolescent health. However, 63%–87% [1, 2] of adolescents obtain less than the 8–10 h of nightly sleep recommended by the National Sleep Foundation [3]. There is evidence that poor sleep habits during adolescence can have negative physiological, cognitive, and behavioral consequences [4, 5], including an increased risk for developing obesity [6]. While the mechanisms underlying the relationship between short sleep and increased obesity risk for adolescents are not known, changes in dietary processes may play a substantial role [7, 8]. Adolescence serves as a unique developmental period for establishing lifestyle behaviors [9, 10], such as healthy sleep (HS) and eating patterns, suggesting that examining the influence of sleep on diet in adolescence is of critical importance.

There is a need to better understand dietary mechanisms by which chronic sleep restriction (SR), manifested as shortened sleep commonly experienced by adolescents on school nights, increases the risk of obesity. We recently observed that adolescents significantly increase the amounts of foods high in glycemic load, carbohydrates, sugar-sweetened beverages, and added sugars following a week of shortened sleep, compared to a week of HS [11]. This, along with findings suggesting a unique increase in the rated appeal of sweets [12] and overall food reward [13] when adolescents were sleep restricted, led us to hypothesize that chronic SR impacts hedonistic response to food, modulating the processing of reward. Our recently published follow-up study [14] in a new cohort of adolescents supported that broad hypothesis: compared to a well-rested state, adolescents experiencing SR rated foods as more appealing and were willing to pay increasingly higher amounts of hypothetical money to obtain their desired foods. However, unlike our prior work, these effects were not stronger for sweet foods than for other types of foods.

Importantly, prior literature did not address the underlying neurological processes by which SR might impact dietary intake. Adult research using functional magnetic resonance imaging (fMRI) suggests that SR increases neural activation in anterior cingulate cortex, insula, amygdala, and other brain regions in response to food-related stimuli [15–19], brain regions that are collectively known to be part of the reward network that drives appetitive responses to food [20, 21]. These findings support our premise that chronically experienced SR can increase food reward in adolescents. However, the prior studies generally involved full-night sleep deprivation, rather than partial-night SR, which is more typically observed during adolescence [22]. Further, adolescent sleep differs from that of adults in important ways, including overall need, sleep architecture, phase timing and amplitude, and the slope by which homeostatic sleep drive accumulates [23]. Adolescents further differ from adults in dietary needs and food perception [24–28]. Finally, known changes in brain processing of rewards and decision-making across adolescence [29–30] complicate generalization from adult fMRI findings. Adult findings can guide hypotheses, but to understand the impact of SR on adolescents’ neural responses to food, one must study adolescents.

To the best of our knowledge, only one study [31] has examined the neurological processes that may drive the increase in food reward observed in sleep-restricted adolescents [12–14, 32]. Jensen et al. [31] recently demonstrated that adolescents undergoing SR exhibit increases in neural activation in the bilateral posterior cingulate cortex, a brain region associated with food reward processing, after viewing food images, compared to undergoing HS. However, this effect no longer reached significance after the removal of two plausible outliers, suggesting that further research on this topic is warranted.

The current study aimed to advance our understanding of the impact of sleep duration on reward processing of food in adolescents, using neuroimaging. We hypothesized that greater activation in reward-related regions in response to visual food stimuli would occur when adolescents experienced SR compared to when they were well-rested. We also explored whether the effect of SR on neuronal responses to visual food stimuli was modulated by various physical and demographic characteristics of the adolescents, changes in the rating of food appeal between sleep conditions, and the experimental order of sleep conditions.

Methods

The Institutional Review Board at Cincinnati’s Children’s Hospital Medical Center approved and monitored all study procedures. Parents and adolescents were provided information about the study, and each provided verbal and written assent/consent to study procedures. Findings from this study are part of a larger experimental trial aimed to examine the impact of experimentally shortened sleep on weight-related behaviors in adolescents. The experimental sleep manipulation and imaging data acquisition occurred over the summer months of 2015–2018 to avoid the negative impact of shortened sleep on academic performance. The experimental manipulation employed in this study is identical to that described previously [14] and briefly outlined in the “Sleep Manipulation” section below.

Participants

This study recruited a sample of healthy adolescents (ages 14–17 years) through online advertisements, community flyers, and e-mails distributed across a large pediatric health network. Research staff initially determined eligibility for all research participants through a phone interview. Exclusion criteria included atypical sleep patterns (regularly sleeping <6 h or >10 h on school nights) or self- or parent-reported symptoms of obstructive sleep apnea or periodic limb movement disorder; standard fMRI contraindications (e.g. metal implants); current psychiatric disorders (i.e. major depression, bipolar disorder, psychosis, eating disorders) screened via validated clinical interview, history of neurological illness or injury (e.g. traumatic brain injury, epilepsy); intellectual disability; use of medications known to alter sleep (e.g. stimulants); daily caffeine consumption that exceed one cup of coffee/energy drink or two caffeinated sodas; and obligations that would preclude adolescents from adhering to the sleep protocol outlined below. Following eligibility determination, adolescents were mailed a sleep diary, actigraph, and detailed study instructions.

Sleep manipulation

Each participant was asked to follow a 3-week sleep protocol [14] shown schematically in Figure 1, A. The first 5 days constituted a stabilization period during which the participants were asked to wake at a time that would allow them to attend 8:00 am research visits; participants held this wake-time constant throughout the 3-week duration of the study, with bedtime being adjusted based on the experimental condition. During the stabilization period, adolescents were allowed to self-select their bedtime. This stabilization approach was chosen to mimic adolescent sleep patterns observed during the school year (i.e. where wake times are predetermined but bedtimes are self-selected). Following the stabilization period, adolescents and their parents attended an 8:00 am Saturday morning office visit, which included in-person review of actigraph data to determine adherence to prescribed wake time. Those with wake times >1 h from their prescribed time were dropped from the study prior to randomization.

Figure 1.

A: The sleep manipulation protocol. SR = sleep-restricted condition. HS = healthy sleep condition. Stabilization, HS, and SR conditions each lasted 5 nights/days. Washout periods covered 2 nights/days each. fMRI and other assessments were performed on Saturday mornings after HS and SR periods. B: Schematic of fMRI paradigm. Time intervals are not shown to scale. Created with BioRender.com.

Following the stabilization period and a 2-night washout period (where wake time was held constant with bedtime being self-selected), eligible participants completed two 5-night sleep periods (SR; 6.5 h/night sleep opportunity) and HS (HS; 9.5 h/night sleep opportunity), created by adjusting bedtime earlier or later in the evening. Using a within-subjects crossover design, each participant was asked to complete both conditions across the course of the study, but the order of conditions randomly varied across participants. Between the two experimental sleep conditions, there was a 2-night washout period where wake time was held constant and bedtime self-selected. On the Saturday morning after each experimental week, participants returned to the lab at 8:00 am in a fasted state (having not eaten breakfast) and were given a standardized meal substitute comprised of an Ensure Original Nutrition Shake of the flavor of their choosing (Vanilla, Chocolate, or Strawberry). Sleep actigraphy data were reviewed to determine adherence to the experimental condition, and study assessments, including fMRI, were performed. All fMRI data acquisition took place between the hours of 8:00 am and 11:00 am, with any given participant undergoing acquisition at approximately the same time across both experimental conditions. During the 3-week duration of the study, all participants were instructed to limit caffeine intake (<1 cup of coffee/energy drink or <2 caffeinated sodas per day), wear actigraphs, complete daily sleep diaries, and refrain from napping.

Background information collected

All participants provided basic demographic information, including age, sex, income, race, and ethnicity. Adolescents’ height and weight were measured in triplicate using research-calibrated scales and stadiometers to determine adolescents’ body mass index (BMI), which was then converted to an age- and sex-adjusted BMI z-score (BMI-z) using the United States Centers for Disease Control and Prevention guidelines [33].

Sleep adherence assessment

Wrist-worn actigraphs (Motionlogger Micro Watch; Ambulatory Monitoring, Inc., Ardlsey, NY) were used to corroborate self-reported sleep diaries that detailed time spent in the bed across assigned sleep conditions. At each Saturday morning assessment, study staff compared the sleep diaries and actigraphy sleep data with adolescents and parents, allowing for the opportunity to reconcile differences between sleep diaries and actigraphy sleep data, identify artifacts in the actigraphy sleep data, and promote adherence to subsequent experimental conditions. After resolving discrepancies, actigraphy-based sleep outcomes were used for the final determination of sleep–wake patterns. Downloaded actigraphy data were scored using a validated algorithm [34] to determine estimates of sleep onset time, sleep offset time, and total sleep duration. Adolescents were deemed adherent to the study if they obtained at least 1 h less sleep during SR than in HS; as previously used in our work [12, 14, 32].

fMRI food paradigm

Each participant underwent neuroimaging while completing a visual food paradigm lasting 15 min and 24 s. A schematic representation of the paradigm is provided in Figure 1, B. At the beginning and end of the paradigm, participants rated their hunger level on a 4-point visual analog scale (“Very hungry,” “Pretty hungry,” “Kind of hungry,” or “Not hungry at all”). The visual food paradigm displayed images to the participants in the MRI scanner using Presentation software (Neurobehavioral Systems, Inc., San Francisco, CA; www.neurobs.com) on an MRI-compatible monitor, with adolescents viewing these images through a mirror attached to the head coil. Image stimuli were drawn from a collection of 210 photographs comprised of 42 photographs of nonfood items (e.g. household items) and 42 photographs of food within each of the following four categories: sweets/desserts (e.g. cookies, candy), processed snacks (e.g. potato chips, pretzels), fast-food entrees (e.g. hamburgers, pizza), and meat/fruit/vegetables. Images were presented in blocks of seven from the same food/nonfood category during the paradigm. Each block of images had a duration of 21 s, displaying each of the seven selected images for 2.5 s, separated by 0.5 s of blank screen. Individual blocks for each category were presented in the order shown in Figure 1, B. This pattern of blocks was repeated six times throughout the paradigm. The order in which the 42 images in each category were presented was randomly shuffled for each run of the paradigm. At the end of each block containing food images, subjects provided a rating of food appeal on a 4-point visual analog scale (“Gross,” “OK,” “Good,” or “Delicious”) using a handheld button box. Similarly, for blocks comprised of nonfood images, subjects provided ratings of likability (“Ugly,” “OK,” “Pretty Nice,” and “Like a lot”). Participants experienced a completely different set of 210 images for their first and second sessions of fMRI.

Imaging acquisition parameters

Scans were performed on a Philips Achieva 3 Tesla MRI scanner (Philips Corporation, Eindhoven, The Netherlands). A T2*-weighted, gradient-echo, echo planner imaging sequence was used for continuous image acquisition during the food appeal paradigm with these parameters: repetition time (TR) = 2000 ms, echo time (TE) = 30 ms, field of view (FOV) = 224 × 224 mm, matrix size = 80 × 80, and 41 axial slices, 3 mm thick, covering the brain volume. A total of 468 image volumes were acquired during the paradigm. In addition, we obtained a T1-weighted, high-resolution 3D structural scan using a Magnetization-Prepared Rapid Acquisition Gradient Echo sequence [35], with TR/TE/inversion time = 8.1/3.7/941 ms, flip angle = 8°, FOV = 256 × 224, 160 slices (spatial resolution = 1 × 1 × 1 mm³), to use as an overlay and anatomic reference for functional parametric images.

Imaging processing

Imaging data were preprocessed using Statistical Parametric Mapping software (SPM12, https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). The first six images of each functional series were excluded to obtain T1 equilibrium before we applied the standard preprocessing pipeline in SPM12 on the remaining functional images, including realignment for motion correction, slice timing correction, anatomical image coregistration to the mean functional image, unified segmentation of the anatomic image, normalization transformation of the anatomic and functional images to Montreal Neurological Institute space, and smoothing of the functional images with an 8 mm Gaussian kernel.

Statistical analysis

Statistical analysis of the preprocessed functional series was performed at the first level, voxel-by-voxel, under a general linear model (GLM) framework including the following regressors: the presentation time course of each stimulus category, the time course of the rating periods, and the three translation and three rotation motion correction time courses. In addition, the Artifact Detection Toolbox (ART, The Gabrieli Lab, McGovern Institute for Brain Research, MIT; www.nitrc.org/projects/artifact_detect/) was utilized to identify outlier volumes exceeding 3 mm in differential composite motion or having intensity variation greater than three standard deviations from the mean. A regressor was added to the model for each outlier volume to scrub their contribution. Finally, we extracted and used as regressors the first five principal components of signal variation within white matter and cerebrospinal fluid tissue classes to further account for nonneuronal contributions that might arise from physiological processes [36].

Contrasts of GLM parameters were considered per participant for each sleep condition for analysis at the second level (random-effects analysis). Primary analysis focused on the contrast of the mean response of all four food categories versus the nonfood stimulus response (food vs. nonfood). A composite activation map, corresponding to the mean food versus nonfood contrast across all subjects and across both sleep conditions, was generated to establish the networks stimulated by our paradigm. The effect of sleep condition on this contrast was determined, within-subject, by pairwise subtraction of the contrast in the two conditions (SR–HS). A linear regression provided the mean sleep condition effect across subjects, adjusted for within-subject change in mean hunger rating during the SR and HS conditions. Secondary analyses generated a map of the effect of sleep condition for each individual type of food versus nonfood contrast.

We also explored potential modulation of the sleep condition effect by various participant characteristics including age, sex, BMI-z, and family income. In addition, we considered potential modulation by order of the experimental sleep conditions experienced by each participant (SR first vs. HS first) and by the change in mean food appeal rating reported by each participant between the SR and HS conditions. We added each modulator individually as a regressor in the linear model while continuing adjustment for change in hunger rating. Voxel clusters in all parameter contrast maps were reported as statistically significant at the cluster level with family-wise error (FWE) corrected p < 0.05, after imposing a voxel level threshold of p < 0.005. Driven by our hypothesis, we also assessed the sleep condition effect restricted to select core regions of the reward network, applying small volume correction after thresholding at p < 0.001 at the voxel level. Cortical and subcortical regions used in small-volume correction were selected from the automated anatomical labeling atlas [37]. Midbrain regions were specifically obtained from the Duke University Adcock Laboratory (https://www.adcocklab.org/neuroimaging-tools), as described in their publications [38, 39].

Results

Participants

A total of 52 adolescents were recruited to participate in the fMRI portion of the larger study; of these, 39 were included in the final analyses (mean age = 16.0 ± 1.1 years, ranging from 14.1 to 17.9 years, 56.4% female). Of those participants excluded, three were nonadherent to the sleep manipulation, one did not return for the final study visit, one did not complete the paradigm, seven failed to record responses during the paradigm (raising concern for equipment malfunction or participant falling asleep), and one was excluded due to a corrupted imaging data transfer. The mean BMI-z of the final sample was 0.19 ± 0.93, and the median family income range was $100K–$125K. Twenty of the participants experienced the SR condition first and 19 experienced the HS condition first.

Sleep behavior and food ratings

Sleep duration was greater for the HS compared to the SR condition by an average of 2.17 ± 0.61 h (p < 0.001). As intended, this was almost exclusively due to earlier sleep onset during the HS condition (averaging 22:27 h ± 45 min for HS vs. 00:43 h ±29 min for SR, p < 0.001). Sleep efficiency (time asleep/sleep period) was well over 85% for both sleep conditions, but higher for the SR (93.79%) condition compared to the HS condition (89.60%) on average as commonly experienced under SR due to homeostatic response. See Table 1 and Figure 2.

Table 1.

Sleep behavior and rating responses according to sleep condition

Metric	HS Mean ± SD	SR Mean ± SD	P (d***)	SR—HS paired Mean ± SD	P (d***)
Sleep onset time (hh:mm)	22:27 ± 00:45	00:43 ± 00:29	<.001 (3.60)	02:16 ± 00:40	<.001 (3.37)
Sleep onset time range	20:52 − 00:25	23:47 − 01:53		00:59 − 04.32
Wake time (24-h)	6.86 ± 0.43	6.96 ± 0.31	.25 (0.26)	0.10 ± 0.44	.17 (0.22)
Sleep duration (hours)	8.42 ± 0.74	6.24 ± 0.51	<.001 (−3.42)	−2.17 ± 0.61	<.001 (−3.58)
Time Asleep (hours)	7.53 ± 0.84	5.85 ± 0.54	<.001 (−2.38)	−1.68 ± 0.60	<.001 (−2.81)
Sleep Efficiency (%)	89.60 ± 5.89	93.79 ± 3.38	<.001 (0.87)	4.19 ± 5.15	<.001 (0.81)
Mean hunger rating during fMRI*	2.35 ± 0.74	2.43 ± 0.86	.64 (0.11)	0.04 ± 0.91	.79 (0.04)
Mean food ratings during fMRI**
All foods	2.78 ± 0.38	2.85 ± 0.35	.44 (0.18)	0.06 ± 0.36	.27 (0.18)
Sweets/desserts	2.85 ± 0.70	2.95 ± 0.63	.51 (0.15)	0.10 ± 0.55	.27 (0.18)
Processed snacks	2.74 ± 0.61	2.79 ± 0.52	.74 (0.08)	0.04 ± 0.43	.54 (0.10)
Fast-food entrees	2.82 ± 0.71	2.91 ± 0.64	.59 (0.12)	0.08 ± 0.65	.43 (0.13)
Meat/fruit/vegetables	2.72 ± 0.51	2.75 ± 0.57	.78 (0.06)	0.03 ± 0.46	.65 (0.07)

^*Hunger ratings: 1. “Very hungry,” 2. “Pretty hungry,” 3. “Kind of hungry,” or 4. “Not hungry at all.”

^**Food ratings: 1. “Gross,” 2. “OK,” 3. “Good,” or 4. “Delicious.”

^***Effect size expressed as Cohen’s d.

SD, standard deviation; SR, sleep-restricted condition; HS, healthy-sleep condition.

Figure 2.

Mean sleep periods in the sleep restricted (SR, light gray) and healthy sleep (HS, dark gray) conditions from sleep onset time to wake time. Bars indicate standard deviation.

One-way ANOVA of mean food rating during fMRI did not detect a difference among food types for the HS condition (F(3, 152) = 0.40, p = 0.75), the SR condition (F(3, 152) = 1.02, p = 0.39), or for the paired difference SR–HS (F(3, 152) = 0.13, p = 0.94). The mean ratings of all foods collectively and of each food type individually during fMRI under the HS and SR conditions did not have discernible differences. Similarly, the mean change in food rating, for all foods collectively or for any food type individually, between sleep conditions within-subject could not be distinguished from zero. Details are provided in Table 1.

Response to viewing food versus nonfood images

The mean contrast of food versus nonfood across all participants and sleep conditions is shown in Figure 3. Viewing food images generated greater response than nonfood images in reward processing regions including bilateral insulae, amygdala, hippocampus, and the anterior cingulate. Strong activation was also seen in the visual cortex, extending into the fusiform gyri. Deactivation (nonfood > food) was observed in bilateral inferior frontal and temporal gyri, extending into middle occipital areas. A more detailed accounting of activated and deactivated regions is provided in Table 2. The mean contrast of food versus nonfood across all participants for the HS and SR conditions separately is provided in Supplementary Figure 1.

Figure 3.

Composite representing the mean contrast of response to viewing food versus nonfood images. Hot colors indicate food > nonfood, while cool colors indicate nonfood > food. All clusters shown are formed by voxels above the threshold of P < .005 and are significant at P < .05, corrected cluster-wise for multiple comparisons. MNI z coordinate is provided above each slice. Neurological orientation convention is used (left side of brain shown on the left side of each image).

Table 2.

Details of voxel cluster characteristics for various activation contrasts and modulations

Contrast	Cluster level P-value* FWE corrected	Cluster size (voxels)	MNI coordinates of peak voxel x, y, z (mm)	Anatomic regions
Composite activation (voxel level p< .005)
Food > nonfood	<0.0001	9825	−12, −87, −16	Bilateral: calcarine, lingual, fusiform, cuneus, precuneus, occipital (sup, mid, inf), midbrain, cerebellum, hippocampus, parahippocampus, insula, amygdala, posterior cingulate, orbital frontal, rectus, olfactory, temporal pole. Left: temporal (inf, sup), operculum.
	<0.0001	1011	28, 48, 35	Bilateral: anterior cingulate, middle cingulate Right: frontal (sup, mid). Left: frontal (sup medial).
	0.0089	239	−27, 48, 41	Left: frontal (sup, mid).
	<0.0001	830	−40, −31, 41	Left: postcentral, parietal (sup, inf), supramarginal.
Nonfood > food	<0.0001	1052	42, −84, 29	Right: temporal (mid, inf), occipital (sup, mid, inf), angular, parietal (sup).
	<0.0001	2172	−45, −84, 26	Left: temporal (sup, mid, inf), occipital (mid), angular, supramarginal, temporal pole, parietal (inf).
	0.0057	259	−31, −42, −10	Left: fusiform, parahippocampus, lingual.
	<0.0001	1062	−54, 22, 5	Left: frontal (inf orbital, inf trigeminal, inf opercular, mid, sup), precentral, orbital frontal, insula.
	0.0001	365	−9, 42, 47	Bilateral: frontal (sup medial), frontal (sup).
	<0.0001	524	56, 31, 11	Right: frontal (inf orbital, inf trigeminal, inf opercular, mid), precentral.
Effect of sleep condition (voxel level P < .005)
SR > HS	0.0072	221	−12, −22, −4	Left: ventral tegmental area, substantia nigra, thalamus, hippocampus, parahippocampus, fusiform, cerebellum.
Effect of sleep condition (voxel level P < .001, using SVC)
SR > HS	0.0030	35	−12, −22, −4	SVC within substantia nigra.

^*Minimum significant cluster size; for composite activation = 167 voxels and for effect of sleep condition = 149 voxels.

FWE, family-wise error; MNI, Montreal Neurological Institute; SVC, small volume correction; inf, inferior; sup, superior; mid, middle.

Effect of sleep condition

Taking the mean difference in food versus nonfood activation between sleep conditions, SR—HS, while adjusting for the mean change in hunger rating between the SR and HS conditions, resulted in identification of a single voxel cluster, shown in Figure 4 and detailed in Table 2, for which activation was greater for SR than for HS. Voxels of the left substantia nigra (SN) and ventral tegmental area (VTA), contributed to this cluster, with extensions into left hippocampus, parahippocampus, and fusiform structures. The effect of sleep condition restricted to core regions of the reward network, including bilateral SN, VTA, hippocampus, amygdala, insula, OFC, and striatal regions, yielded a cluster within the SN region significant with p < 0.05, FWE corrected, after applying a voxel level threshold of p < 0.001 uncorrected (see Table 2).

Figure 4.

The main effect of sleep condition (mean within-subject difference SR—HS) on brain activation resulting from viewing food versus nonfood images, with adjustment for the mean change in hunger rating between the SR and HS conditions. Hot colors indicate positive contrast. The cluster shown was formed by voxels above the threshold of p < .005 and significant at p < .05, corrected cluster-wise for multiple comparisons. MNI z coordinate is provided above each slice. Neurological orientation convention is used (left side of brain shown on the left side of each image).

Contrasts of individual food types versus nonfood did not result in significant clusters, with the exception of snack foods, which had a voxel cluster for which activation was greater for the SR versus HS condition. This cluster comprised regions in common with the cluster generated by the food versus nonfood contrast. This cluster is shown in Supplementary Figure 2 and detailed in Supplementary Table 1.

Modulation of the sleep condition effect

The sleep effect for food versus nonfood was found to be linearly related to two factors. A negative relationship with the sleep condition effect was found for family income in temporal regions extending to the angular gyri, inferior frontal lobe, insula, and cerebellum (Supplementary Figure 3, A). The change in mean food appeal rating between the SR and HS conditions had a negative relationship with the sleep condition effect in widespread frontal and temporal brain areas and included a central cluster comprised of the SN and VTA extending to occipital regions (Supplementary Figure 3B). Details of the clusters shown in Supplementary Figure 3 are provided in Supplementary Table 2. The sleep condition effect was not found to be significantly modulated by sex, age, BMI-z, or sleep condition order.

Discussion

In this study, viewing images of food was found to elicit robust neuronal responses including many canonical reward-processing regions of the brain in healthy adolescents. SR, to a degree commonly experienced by adolescents today, resulted in greater activation of central components of the dopamine-driven mesolimbic reward processing network including the VTA and SN. This SR effect on brain response was observed for all food types collectively and for snack foods individually, though mean ratings of food appeal were not substantively altered by the SR imposed in this imaging subsample from our larger study. Exploratory analyses showed that some participant characteristics correlated regionally with the SR effect.

Extensive research into the reward processing network has been summarized in several integrative reviews [20, 21, 40]. The network includes four highly-connected core regions: the amygdala/hippocampus, striatum, insula, and OFC. Encoding of incentive value [41], integration of sensory inputs [42, 43], learning and memory of food value [44], and linkage of motivation to action [45] are performed by these core elements and modulated by hunger level [46, 47]. Higher-level cognitive control cortices, such as inferior and lateral prefrontal cortices and the anterior cingulate cortex play evaluative and inhibitory roles that modulate motivation, action, and the incentive value of food cues [48]. Central to the network are the VTA and SN, midbrain structures that project anticipatory dopamine (DA) signaling to other components of the network on exposure to food cues. DA signaling influences the entire network, providing an index of reward value and motivation [49–52].

Our results for the mean response to food pictures across participants and sleep conditions included significant activation in each of the reward network core regions as well as in cognitive control areas of the frontal lobe. Absent were detectable responses in the VTA and SN brainstem structures when averaging across both sleep conditions. Pairwise differences in activation between the SR and HS sleep conditions, however, highlighted exclusively a cluster that included the left VTA and SN, extending into surrounding parahippocampus, hippocampus, and fusiform structures, where food versus nonfood contrast was greater for the SR compared to the HS sleep condition. Behaviorally, there is evidence from prior studies [32, 53] that SR or total sleep deprivation leads to increased appetite and food intake in adolescents. Our recent research has demonstrated that there is a particular influence of adolescent sleep on the intake of foods high in glycemic index [11]. Prior adult neuroimaging studies of the impact of poor sleep on the brain’s response to food-related stimuli have shown that various elements of the reward network increase activation under poor sleep conditions [15–19]. Imaging studies in adolescents have indicated weakened inhibitory responses in individuals exposed to highly motivating food cues [54]. SR was also found to diminish inhibitory brain activation to food cues in adolescents with obesity versus normal weight [31]. These prior studies are often limited to small sections of the network or are not consistent regarding the regions impacted by sleep manipulation. This may be due to low sample sizes or to differences in study populations, food-related stimulation paradigms, or study designs. Our study did not discern differences between SR and HS in several aspects of core reward circuitry, as observed to various degrees in other imaging studies. This may be due, in part, to our use of a less intense chronic partial SR protocol, rather than all-night sleep deprivation. We also speculate, however, that our study design focused on reward signaling, specifically the DA-driven reward-prediction error signal provided by the VTA and SN, where we observed marked increases for the SR condition. The core reward network, on the other hand, is influenced by more stable factors like the learned association of food and food cues with nutritive and hedonic value that underlie the function of these core regions [20].

Our prior work has yielded mixed results on whether specific types of foods are perceived differently during SR than when adolescents are well-rested; one study showed a specific effect on sweet/dessert foods [12], but subsequent findings suggested effects on other types of foods as well [11, 13]. We explored this possibility by systematically including sets of stimuli specific to sweets, fast-food entrees, snacks, and foods consumed at everyday meals such as meats, fruits, and vegetables. Snack foods individually elicited a sleep effect in central and posterior structures in common with those observed for all foods collectively (Supplementary Table 1 and Supplementary Figure 2). This observation motivates the hypothesis that calorie-rich, high-fat food categories, such as snacks, are especially incentivized by SR through increased dopaminergic activation in the midbrain. While there is some consistency across previous studies that SR selectively increases intake of calorie-rich, nutrient-poor foods, imaging studies have not consistently found appetitive responses to SR to be uniquely strong in those foods [15, 16, 31]. As such, the mechanism underlying this selective increase in intake following SR remains unclear. Given that calorie-rich, nutrient-poor foods are plentiful in the developed world, it remains possible that increased intake during SR is related to diminished inhibitory control, even if their appeal remains constant [7].

Assessment of the moderating effect of various participant characteristics on the sleep effect (contrast between SR and HS conditions for food vs. nonfood activation) (Supplementary Figure 3 and Table 2) yielded several reward regions that were activated by the food paradigm (see Figure 2). Higher family income appears to diminish the SR versus HS response in temporal and cerebellar regions (Supplementary Figure 3, A). The possibility that higher socioeconomic status may mitigate the impact of SR on reward network function should be investigated further.

Moderation of the change in response during SR versus HS by the change in mean food appeal rating between SR and HS was also found to be negative in a broad distribution of regions, including the SN and surrounding structures where we observed a significant increase in response to food for SR versus HS. This result is surprising since it suggests that augmentation of dopaminergic drive under SR is diminished as adolescents increase how much they like the viewed foods. Response to food reward can be driven by two intertwined but separate processes, the “wanting” of food versus the “liking” of food [49]. The “wanting”, or incentive salience, is elicited by food cues and refers to motivation to seek and obtain reward. It is mediated by the dopaminergic mesolimbic network centered on the VTA and SN. “Liking” refers to hedonic appeal of food tasted, learned and stored in memory, measured in this study by the subjective food ratings. Hedonic appeal is linked to ventral forebrain including the OFC and nucleus accumbens. While “liking” and “wanting” are certainly linked, they are considered separate concepts on which SR may have an opposing impact. While these collective results reveal some moderation of the impact of SR by various experimental and participant-specific factors, they must be considered exploratory and suggestive of hypotheses worth pursuing in future work.

Statistical power considerations

A strength of our experimental design is the imposition of SR that is representative of the adolescent experience. Nevertheless, the effect of this sleep manipulation is mild compared to total sleep deprivation protocols. There is utility, therefore, in reporting effect sizes of our results to inform sample size specification of future experiments with similar SR protocols. Table 1 includes effect sizes for sleep metrics and food ratings in the HS and SR conditions. Effect sizes for food ratings are small, with values of 0.18 or less. Such effect sizes would dictate samples sizes of well over 200 subjects for discernment of differences at p < 0.05 with power of 80%. Pairwise differences in brain response to the food paradigm between sleep conditions, at least at the voxel level, include some regional effect size absolute values exceeding 0.3, as shown in Figure 5. Indeed, some of these regions, including insula (SR > HS) and medial prefrontal cortex (HS > SR) are outside the significant cluster reported in Figure 4. The effect size map also suggests that the left-sided significant cluster of Figure 4 could extend bilaterally in a study with a larger sample size. An effect size of 0.3 for this pairwise comparison would suggest sample size of 87 at p < 0.05 and 80% power, a reasonable specification for a future study. It should be noted that this does not account for multiple comparisons or cluster size for brain-wide analysis but may at least provide targets for region-of-interest ROI analyses.

Figure 5.

Effect size (Cohen’s d) for the mean within-subject difference (SR—HS) of brain activation resulting from viewing food versus nonfood images. Hot colors indicate positive contrast while cool colors indicate negative contrast. The clusters shown were formed by voxels above the threshold of Cohen’s |d| > 0.3. MNI z coordinate is provided above each slice. Neurological orientation convention is used (left side of brain shown on the left side of each image).

Limitations

Outcomes of this study should be interpreted with consideration of some limiting factors. Although our sample size was larger than prior studies of behavioral and neuronal appetitive changes induced by sleep manipulation, the sleep manipulation utilized was modest, raising the possibility that our sample size lacked power to detect the correspondingly small effect sizes as described in the previous section. However, our sleep manipulation reflected realistic sleep experiences of our target adolescent population (e.g. late self-selected bedtimes, early pre-determined rise times); results from more extreme sleep deprivation paradigms may not reflect what happens in daily life for these youth. Our study did not discern a moderation of the sleep effect by order of sleep condition. Nevertheless, there is a risk for a potential artifact of the study design. Although the initial sleep stabilization period and washout periods were not intended to induce SR, participants averaged ~7 h of sleep those nights, which is below consensus recommendations of 8–10 h. As a result, the effective period of SR was longer when SR preceded HS than when HS was experienced first. Reassuringly, a direct comparison, by two-sample t-test, of the sleep effect in those experiencing SR first (n = 20) versus those experiencing HS first (n = 19) yielded no significant clusters of voxels. It is also notable that only the visual aspects of food cues were presented to the participants, neglecting other potentially powerful drivers of appetitive response, including gustatory and olfactory stimulation. Composite activation for the food versus nonfood contrast (Figure 3) was markedly strong in primary visual cortices. This may be due to imperfect matching of image color and complexity between food and nonfood images. Finally, the employed food paradigm was passive and not designed specifically to assess inhibitory processes. The inhibitory response is part of the cognitive control aspect of reward processing that could, in theory, act as a countermeasure to increased appetitive drive that is vulnerable to weakening by SR. Responses to the visual paradigm, however, do not exclude inhibition, as evidenced by composite frontal lobe responses.

Conclusion

The results of this study suggest that SR, at the level commonly experienced by healthy adolescents, resulted in an elevated dopaminergic drive of the mesolimbic reward brain network. This may lead to augmented motivation to seek food in the context of each individual’s underlying profile of food appeal and inhibitory mechanisms. In an environment in which foods, particularly heavily marketed calorie-rich, nutrient-poor foods, is abundantly available, augmented incentive salience by SR can potentially lead to overeating. Present data suggest that countermeasures that reduce food salience could include promoting consistent HS durations, and perhaps controlling the availability of unhealthy foods during times when sleep routinely dips (e.g. final exam studying).

Supplementary material

Supplementary material is available at SLEEP online.

Disclosure Statement

Financial Disclosure: none.

Non-financial Disclosure: none.

Data Availability

The data underlying this article will be shared on reasonable request to the corresponding author.