Increased brain and behavioural susceptibility to portion size in children with loss of control eating

Summary

Background

Portion size influences intake (i.e. the portion size effect [PSE]), yet determinants of susceptibility to the PSE are unclear.

Objective

We tested whether children who reported an episode of loss of control (LOC) eating over the last 3 months would be more susceptible to the PSE and would show differential brain responses to food cues compared with children with no-LOC.

Methods

Across five sessions, children (n = 47; 7–10 years) consumed four test meals at 100%, 133%, 167% and 200% conditions for portion size and completed a functional magnetic resonance imaging scan while viewing pictures of foods varied by portion size and energy density (ED). Incidence of LOC over the past 3 months was self-reported. Random coefficient models were tested for differences in the shape of the PSE curve by LOC status. A whole-brain analysis was conducted to determine response to food cues during the functional magnetic resonance imaging.

Results

Reported LOC (n = 13) compared with no-LOC (n = 34) was associated with increased susceptibility to the PSE, as evidenced by a positive association with the linear slope (P < 0.005), and negative association with the quadratic slope (P < 0.05) of the intake curve. Children who reported LOC compared with no-LOC showed increased activation in the left cerebellum to small relative to large portions (P < 0.01) and right cerebellum to High-ED relative to Low-ED food cues (P < 0.01).

Conclusion

Children who reported LOC were more susceptible to the PSE and showed alterations in food-cue processing in the cerebellum, a hindbrain region implicated in satiety signalling.

Introduction

Food portion size influences intake, but why the ‘portion size effect’ (PSE) occurs is unclear. Results from studies in children show that both food portion size and energy density (ED; kcal g⁻¹) have independent and combined effects on energy intake (1–4). The most robust effect of portion size on intake is seen for high energy-dense foods (5), and this is a concern for the development of childhood obesity. However, the mechanisms underlying the effects of portion size and ED are not clear. Cross-sectional (6) and epidemiological studies in children (7) suggest a positive association between portion size of food consumed and obesity. In addition to weight status (8), other individual characteristics such as age (1,5,9), appetitive traits (2,10) or disinhibited eating (11) can influence susceptibility to the PSE. Understanding which factors may predispose some children to overeating from large portions, of high-ED foods in particular, is an essential step towards understanding individual differences in the susceptibility for obesity. The goal of this study was to investigate whether loss of control (LOC) eating in children is associated with increased susceptibility to the PSE and altered brain responses to food cues that vary by portion size and ED.

Eating-related LOC, herein ‘LOC’, is characterized by the inability to control what or how much is being consumed, regardless of whether the amount consumed is perceived to be large or not. Therefore, LOC itself differs from classic binge-eating episodes, which include both the perceived LOC and the consumption of objectively large amounts of food (12). Reporting of LOC is estimated in approximately 4% (13) of children and 33% (14) of adolescents. LOC often emerges during middle childhood (12,15), remains moderately stable over 5 years (16) and is positively associated with adiposity (17), excess weight and fat gain (18–20) and other forms of disinhibited eating (17,21). LOC may occur more commonly in girls than boys (22) and is persistent in half of children who report it (16,23). Although not a clinical eating disorder, LOC appears to be a marker of adverse eating and weight-related outcomes. A few studies have examined the association between LOC and objectively measured energy intake (24–29). Most of these studies tested consumption across two sessions where participants were instructed to ‘let themselves go’ at one session and eat normally at another (25,26). Those with LOC compared with without had greater self-reported energy intake from high-ED foods (30) and consumed more high-ED foods from laboratory meals (28–30). However, total energy intake at binge-like meals did not vary between those with LOC compared with without (27,29).

Investigating the neural underpinnings of food-cue processing may shed light on the origins of LOC eating in children. Little is known about the brain correlates of LOC eating; however, examination of the broader literature on food cue responding in youth with and without disordered eating may inform the current study. For example, in adolescents without eating disorders, viewing high-ED food cues is associated with heightened engagement of fronto-striato-limbic circuitry (e.g. regions believed to operate in decision-making, reward and emotional processing) (31,32). In contrast, adolescents at risk for obesity with greater emotional eating had reduced frontolimbic activation after receipt of a high-ED milkshake (33). Evidence in adolescents and adults with bulimia nervosa supports that abnormal maturation of frontostriatal systems may contribute to the development of disordered eating (34). Following this, one recent study found that reduced engagement of prefrontal brain regions may be associated with overeating in adolescent youth (mean age = 15.8 years) with LOC (35). However, there is a need to better understand how reported LOC eating relates to brain and behavioural responses to food cues in middle childhood, when LOC eating is likely to emerge.

The purpose of this study was to determine whether LOC eating in children is associated with susceptibility to the PSE and to characterize the underlying neurobiological mechanisms for this susceptibility. We hypothesized that children who reported LOC compared with no-LOC would show greater susceptibility to the PSE across four multi-item test meals, particularly for higher ED foods. Although we conducted whole-brain analyses to generate hypotheses, we also anticipated that children who reported LOC compared with no-LOC would show decreased activation in inhibitory control regions and increased activation in reward-based regions when contrasting food cues by portion size (e.g. large vs. small) and ED (e.g. higher-ED vs. lower-ED).

Methods

Design

The current analysis addressed a secondary hypothesis from a larger clinical trial, which was designed to determine brain mechanisms underlying the PSE using functional magnetic resonance imaging (fMRI) (36–38). This study used a crossover design with LOC eating treated as a between-subjects factor and portion size as a within-subjects factor. Enrolled participants attended five total visits. Children came to the laboratory with their parents for each visit. Children were served a meal consisting of six age-appropriate and commonly accepted foods on four visits conducted at approximately the same time each week. Visits were scheduled at either lunch or dinner time depending on family availability but kept at consistent times within families. Parents were instructed to have children refrain from eating or drinking at least 2 h before each visit. Across the four meals, participants were served either the reference (100%) portion of all foods, 133%, 167% or 200% of the reference amounts. The order in which portion size conditions were delivered was randomized and counterbalanced across subjects. On the last visit, children viewed images of foods that varied by ED (>1.5 or <1.5 kcal g⁻¹) and portion size (large and small) during an fMRI scan. For the fMRI, we selected a moderate ED of 1.5 kcal g⁻¹ as a cut-off to separate higher from lower energy-dense foods and to allow us to balance food types (e.g. entrées and side dishes) and taste attributes (e.g. sweet and savoury) across the groups. We then designed the test meal with similar categorizations of higher and lower ED foods. Average ED for the pictures shown in the fMRI was 3.4 and 0.6 kcal g⁻¹ for the higher and lower ED foods, respectively. For brevity, herein, we will refer to foods served at the meal as ‘high-ED’ and ‘low-ED’ and foods shown in the fMRI scanner as High-ED and Low-ED. Data collection occurred between July 2013 and July 2015.

Participants

Children aged 7–10 years and their parents were recruited using flyers distributed in the community, postings on university websites and social media and word of mouth. Respondents were screened by telephone to determine whether they met the following criteria: native English speakers, right handed, reading at or above grade level, comfortable in small spaces and willing to eat the foods served in the study. Potential participants were excluded if they had food allergies, metal in or on their body that could not be removed and/or use of medications known to affect appetite and/or brain activity. Parents signed informed consent, and children gave verbal and written assent. Participants were financially compensated at the end of each visit. The Office for Research Protections of The Pennsylvania State University approved the study.

Assessment of child height and weight

Anthropometrics were collected by a trained researcher from participants wearing light clothing, with shoes removed, on the first visit. Height, weight and body fat percentage were measured twice using a stadiometer (Seca North America, Chino, CA, USA), standard scale (Detecto model 437, Detecto, Webb City, MO, USA) and bioelectrical impedance analysis (Tanita model BF-350, Tanita, Arlington Heights, IL, USA). Markers of weight status (body mass index [BMI], BMI z-score and BMI percentile) were calculated according to the Centers for Disease Control and Prevention conversion programme (39).

Assessment of physical activity

To estimate activity level, children wore an ActiGraph GT3X-BT (ActiGraph, LLC, Pensacola, FL, USA) accelerometer on their non-dominant wrist for 7 d for 24 h per day (except when in water). The ActiGraph was placed on the child’s left hand following the second visit, and they returned the watch the following week. Total sedentary time and time spent in moderate-to-vigorous physical activity (MVPA) were extracted for analyses to determine whether children who reported LOC differed from children who did not. Data were scored in ACTILIFE 6 software (ActiGraph, LLC, Pensacola, FL, USA).

Assessment of child eating styles

To assess trait-like eating styles in children, parents answered the 35-item Child Eating Behaviour Questionnaire (CEBQ) (40) on the first visit based on a scale of 1–5 (never–always) for how often their child exhibits various eating behaviours. The 35-item CEBQ reduces to eight sub-scales of traits: satiety responsiveness, food responsiveness, enjoyment of food, emotional overeating, slowness in eating, food fussiness, emotional undereating and desire to drink. Examples of items and corresponding traits include ‘my child has a big appetite’ (satiety responsiveness), ‘if allowed, my child would eat too much’ (food responsiveness), ‘my child looks forward to meal times’ (enjoyment of food), ‘my child eats more when worried’ (emotional overeating), ‘my child eats slowly’ (slowness in eating), ‘my child refuses new foods’ (food fussiness), ‘my child eats less when angry’ (emotional undereating) and ‘my child is always asking for a drink’ (desire to drink). Psychometric properties of the CEBQ include good test–retest reliability, concurrent validity with observed eating behaviour and stability over time (40,41).

Assessment of loss of control eating

The LOC screening questionnaire was developed based on the results of a study utilizing the Paediatric Eating Episode Interview (42). The questionnaire was administered in an interview format by a trained researcher. Children were asked whether or not they experienced an episode in the past 3 months where they felt they could not control the type or amount of food that they ate. Previous data establishes that (i) regardless of episode size, reporting an episode of LOC is the most salient feature of disordered-eating symptomatology (22) and (ii) regardless of episode frequency, LOC is related to adverse health outcomes (17). As such, participants were categorized into LOC or no-LOC groups based on whether they answered yes or no to experiencing at least one episode of LOC in that past 3 months. If the child answered yes, they were prompted to continue answering 19 questions in regard to before, during and after the episode of LOC eating.

Examples of the 19 follow-up questions include Before the episode, did something good/bad happen to make you want to eat?; During the episode, did the amount of food feel like too much for you at the time?; After the episode, did you feel badly about yourself for eating or about what you ate? Children were given the option of skipping questions they felt uncomfortable answering. Three of the 54 children enrolled in the study chose to skip the LOC assessment, so their data were excluded from analyses. Immediately after the assessment, a script was read to all children who reported yes to LOC over the past 3 months to thank them for their honesty and encourage them to speak to a trusted adult. In addition, we also asked the child’s permission to speak with his/her parents about these concerns, although children were allowed to deny this request.

Test meals

The meals were prepared in a research kitchen according to a standardized protocol. Test meals included six foods that were commonly consumed by children of this age (2,3,11) and were chosen based on experience from similar laboratory paradigms (11). The selected foods provided a range of ED but were grouped into low-ED (broccoli, cherry tomatoes and red grapes; ED range: 0.2–0.7 kcal g⁻¹) and high-ED (garlic bread, macaroni and cheese and angel-food cake; ED range: 1.5–3.6 kcal g⁻¹) food categories for practicality in exploring hypotheses. The portion sizes of the reference condition (100%) were chosen based on data from the Continuing Survey of Food Intakes by Individuals (43) and on previous laboratory meal paradigms (11,44). To obtain the other meal conditions, the portion sizes of all foods were simultaneously increased by weight to 133%, 167% and 200% of reference. All foods at each meal were weighed prior to being served and reweighed to the nearest 0.1 g after the child indicated they were finished. Water was provided with the meal ad libitum. Information from food manufacturers (i.e. nutrition facts label) was used to convert the weight consumed to kcal for assessing energy intake. We defined ‘plate cleaners’ as children who finished greater than 95% of the total meal at any condition (N = 1). Removal of this child’s data from the analyses did not influence the outcomes.

Participants rated their fullness before and after each meal, using a vertical, visual analogue scale in the shape of a doll (45). Participants answered the question, ‘How full do you feel right now?’ by using a slider that moved up and down the doll’s stomach. Liking and wanting of each food item at the meal was assessed by having the child taste a small sample (~10 g) of each food and then rate on horizontal visual analogue scales anchored by ‘Not at all’ on the left and ‘Very much’ on the right. Post-meal, participants again rated fullness, liking and wanting using the same measures.

Participants were served their test meals in an observation room with a child-friendly rug, non-food photos on the wall and table with three chairs. Doors were kept ajar to the waiting area, where parents waited throughout the visits. Children were given 30 min to eat as much or as little as they liked. During the test meal, a trained researcher read an age-appropriate, non-food-related book selected by the child to provide a neutral distraction as the child ate at the table.

Functional magnetic resonance imaging testing protocol

The fMRI stimuli, design and parameters have been described previously (37,38). Prior to the fMRI scan, children had been trained utilizing a mock scanner to ensure comfort and data quality during the fMRI visit (36). During the fMRI scan, visual stimuli were shown via projector in a block design with two control conditions of non-food items (i.e. furniture and blurred to control for low level visual characteristics across conditions) and four conditions of food at two levels of portion size and two levels of ED. The categories of food images shown were large-portion High-ED, small-portion High-ED, large-portion Low-ED and small-portion Low-ED. Examples of foods shown in the High-ED category include pasta with cheese sauce and chocolate candies. Examples of foods shown in the Low-ED category include plain brown rice and blueberries. All foods were shown in large-portion and small-portion sizes based on amounts commonly consumed by this age group from the Continuing Survey of Food Intakes by Individuals (43). A complete set of images used including food and control conditions in the paradigm for this study has been previously published (38). Food images were similar in resolution, overall size, brightness and intensity. Although visual food cues lack other sensory aspects of foods (e.g. texture and smell), evidence supports the validity of their use in fMRI as responses to them are reproducible (46), linked with food intake (47) and associated with post-ingestive consequences (48). Following the fMRI, children rated liking and wanting for all of the stimuli on 150-mm visual analogue scales.

Functional magnetic resonance imaging data acquisition

Children entered the fMRI room with a research assistant and MRI technician after parents completed a safety-screening questionnaire. Cushions were used to enhance participant comfort and limit movement during the scan. Structural and functional brain data were obtained from a Siemens MAGNETOM Trio 3T MRI scanner (Siemens Medical Solutions, Erlangen, Germany). Structural brain data were collected via one anatomical scan and functional brain activity via six functional runs, each presenting six blocks of stimuli with five images from a single condition (e.g. large-portion high-ED) per block. After the initial 36 participants were tested, the fMRI facility upgraded the scanner equipment from a standard 12-channel head coil to a Prisma Fit 20-channel head coil. The last 11 participants were scanned on the upgraded system, but all scan parameters were identical pre-upgrade and post-upgrade (37). The imaging centre independently conducted fMRI scans in 31 adults measured before and after the update to determine consistency in responses to a well-validated sensory task. Functional data obtained for this task were similar between pre-update and post-update, although the signal-to-noise ratio (SNR) of the scanner increased. To account for these differences in the present study, we adjusted all models for SNR.

Structural data were acquired via high-resolution T1-weighted scans that used a magnetization-prepared rapid gradient-echo sequence (repetition time (TR): 1650 ms; echo time (TE): 2.03 ms; flip angle: 9°; field of view: 256 mm; slice thickness: 1 mm; sagittal plane, voxel size: 1 × 1 × 1 mm³). Data collected from functional scans were acquired via single-shot gradient T2*-weighted echo planar-imaging with blood oxygenation level dependent (BOLD) contrast. We used the following TR: 2000 ms; TE: 25 ms; flip angle: 90°; and matrix: 64 × 64 with an in-plane resolution of 3 × 3 × 3 mm (field of view: 220 mm) to acquire each volume, which was composed of 33 3 mm interleaved slices along the anterior commissure–posterior commissure plane. To assist with the motion-induced effects during acquisition, in-scan prospective movement correction adjusted slice positioning during scans (49). Each presentation of visual stimuli (i.e. each functional run) was started manually to coordinate with MRI data acquisition. This allowed researchers to verbally confirm participant comfort and alertness in between functional runs.

Data analysis

Behavioural analyses

Data were analysed with SPSS v22.0 software (IBM Corp). Two participants were excluded from analyses because they moved out of state during the middle of the study and missed weeks 3, 4 and 5. Three youth were excluded because they refused to complete the LOC questionnaire. There were no differences in primary outcome variables between children who completed the LOC and those who did not. Therefore, intake analyses included data from 47 children, approximately 20% of whom were overweight or obese according to BMI percentile (Table 1). Distribution plots, summary statistics and the Shapiro–Wilk test confirmed normal distribution of intake in those who reported LOC and no-LOC (data not shown). Characteristics of foods served at the test meals are in Table 2.

Table 1.

Participant characteristics

Characteristic†	No-LOC, n = 34	LOC, n = 13	Range, n = 47
Age (years)	8.9 ± 0.2	8.7 ± 0.3	7.2–10.9
Weight (lb)	71.0 ± 3.2	69.0 ± 5.2	38.6–122.8
Height (in.)	53.7 ± 0.6	52.6 ± 1.0	43.6–60.0
Body fat %	18.2 ± 1.3	18.8 ± 1.9	5.3–38.7
BMI z-score	0.1 ± 0.2	0.3 ± 0.2	−1.5–2
BMI percentile	53.1 ± 4.9	58.0 ± 7.3	6.1–97.9
Average MVPA‡ (min d−1)	207.5 ± 50.5	194.3 ± 38.1	105.4–315
Average sedentary time (min d−1)	376.9 ± 61.1	369.1 ± 68.8	187.6–455.8
	n (%)	n (%)	n (%)
Sex§
Boys	14 (40)	9 (67)	23 (49)
Girls	20 (60)	4 (33)	24 (51)

Values are mean ± standard error of the mean unless noted otherwise.

^†There were no differences between LOC and no-LOC on any of these variables.

^‡Moderate-to-vigorous physical activity scored by ActiGraph.

^§More boys than girls reported LOC; thus, sex was included as a covariate in all models.

BMI, body mass index; LOC, loss of control; MVPA, moderate-to-vigorous physical activity.

Table 2.

Foods served across portion size conditions

		100% portion		133% portion		167% portion		200% portion
Food served	ED	g	kcal	g	kcal	g	kcal	g	kcal
Mac and cheese†	1.4	300	426	400	567	500	711	600	852
Garlic bread (in.)‡	3.6	50 (2)	180	67 (2.7)	239	84 (3.3)	301	100 (4)	360
Angel food cake§	2.5	40	100	53	133	67	167	80	200
Steamed broccoli¶	0.4	90	40	120	53	150	67	180	80
Cherry tomatoes	0.2	50	9	67	12	84	15	100	18
Red grapes	0.7	100	69	133	92	167	115	200	138
Meal weight and kcal		630	824	840	1096	1052	1376	1260	1648

^†Stouffer’s macaroni & cheese (Nestle), Solon, OH, USA.

^‡Pepperidge Farm Inc., Norwalk, CT, USA; weight of garlic bread is approximate as it was cut by inches.

^§Sara Lee, Downers Grove, IL, USA.

^¶Birds Eye Frozen Florets (Pinnacle), Parsippany, NJ, USA.

ED, energy density.

Previous studies that measured energy intake at laboratory meals have observed non-linearity in responding to portion size variations; i.e. the PSE is curvilinear when intake is measured across four or more conditions (44). Furthermore, the magnitude of the PSE can vary depending on how many portion conditions are served (8,11,50). Therefore, polynomial growth curves were used to account for non-linearity (51). This type of function allows the intake curve to peak then decline, which may be true when the portion size of all foods in multi-item meals is increased due to competition between foods.

Random coefficient models were first used to obtain linear and quadratic coefficients for mean weight of food (i.e. gram) and energy (i.e. kcal) consumed across the full sample in response to amount served (i.e. gram weight served at the 100%, 133%, 167% or 200% conditions). Similar models were used to compare differences in the secondary outcomes of intake from high-ED and low-ED foods, separately. Models were developed by adding the intercept, linear and polynomial factors of the amount served as both fixed and random effects. Curves were centred at the reference portion condition. Therefore, the intercept of the curve represents intake of the reference portion condition. The coefficients represent the rate of change in intake (linear) or the deceleration in rate of intake (quadratic) as the portions were increased beyond the reference condition. Models with BMI z-score, child age, study week, time of day tested, MVPA from the ActiGraph, pre-meal fullness ratings and liking of the test-meal foods as covariates were tested, by adding each to the model individually. None of the covariates significantly influenced the models and therefore were dropped from final analyses (data not shown).

Additional participant characteristics (i.e. LOC status) were then added to the random coefficient models as covariates to test whether they accounted for individual differences in intake in response to amount served. Because males were over-represented in the LOC group, sex was retained as a covariate in all models. Modelling in this way allowed us to analyse the trajectories of intake in response to larger portion conditions (i.e. PSE curves) by comparing PSE curves between those who reported LOC and no-LOC. Model-based means ± standard error of the mean are reported and were significant based on a P < 0.05 cut-off.

Functional magnetic resonance imaging data analyses

For the fMRI data, BRAINVOYAGER software, v. 20.6 (Brain Innovation, Maastricht, The Netherlands) was used for whole-brain analyses. Anticipated brain regions of activation included frontal and reward-based regions such as the ventromedial prefrontal cortex, orbitofrontal cortex and striatum, which have previously shown to differential activation in those with LOC compared with without LOC (35).

Preprocessing of fMRI data was conducted with a standard pipeline of steps using BRAINVOYAGER software. Structural data were skull-stripped to reduce the search area and normalized to Montreal Neurological Institute space (52). We justified use of common stereotaxic space based on reports that children older than 6 years have minimal differences in brain anatomy to adults (53), although the benefits of using child specific templates have been acknowledged (54). Preprocessing of the functional data in BRAINVOYAGER included 3-d motion correction using six vectors (three translations and three rotations) and temporal high-pass filtering using a general linear model Fourier basis set with 3 cycles per time course. Finally, data were smoothed prior to entering the group analysis with a 6-mm³ full width at half maximum Gaussian filter. To be included in fMRI analyses, participants had to have a successful structural scan and at least one functional run with <3 mm or 3° of motion in any direction. This left n = 48 in fMRI analyses, with an average of 5.1 successful runs per child (range 3–6).

A whole-brain analysis was conducted in BRAINVOYAGER software to examine whether there were differences in the brain response to food cues between children with and without LOC using a random effects general linear model. Regressors were created for the blocks of food images (large-portion High-ED, small-portion High-ED, large-portion Low-ED and small-portion Low-ED) and entered into a general linear model to obtain variable estimates (i.e. beta weights) for each participant. Those estimates were entered into a second-level ANOVA with the within-subjects factors of ED and portion size and between-subjects factor of LOC status. To correct for multiple testing, results were thresholded with a spatial-extent threshold of 6 (k = 6) and voxel-wise P value of <0.001, which was determined by 10 000 Monte Carlo simulations using the standard procedures of the ClusterThresh plug-in of the BRAINVOYAGER software (Brain Innovation) and applied to statistical maps to yield a corrected alpha of 0.05. Peak Montreal Neurological Institute coordinates from brain regions that exhibited significant effects at P < 0.001 level are presented in Table 3. Post hoc analyses were conducted using Tukey’s Honest Significant Difference test. For exploratory purposes, analyses that remained significant at the P < 0.005 level are provided in Table S1.

Table 3.

Peak coordinates of functional areas from whole-brain analyses with interactions for portion size, energy density and LOC

ROI name	No. voxels	x	y	z	Main effect portion size	Main effect energy density	LOC × portion size	LOC × energy density
Main effects: portion size
L. lingual gyrus	704	−4	−81	−8	37.76**	–	–	–
L. superior temporal gyrus	6	−54	−41	13	15.34**	–	–	–
Main effects: energy density
R. fusiform gyrus	25	45	−57	−17	–	17.46**	–	–
R. fusiform gyrus	8	38	−65	−5	–	14.53**	–	–
R. cerebellar declive	14	23	−70	−17	–	21.00**	–	–
L. medial frontal gyrus	14	−7	56	10	–	17.01**	–	–
L. superior temporal gyrus	12	−36	24	−33	–	17.75**	–	–
Contrast: LOC × portion size
L. cerebellum	33	−46	−54	−39	–	–	22.84**	–
Contrast: LOC × energy density
R. cerebellum	152	6	−52	−28	–	–	–	21.05**

Voxelwise threshold: P = 0.001; cluster threshold: 6.

^**P < 0.01; corrected results are significant.

L, left; LOC, loss of control; R, right: ROI, region of interest.

Following the whole-brain analyses, we exported beta values for the regions that exhibited significant main effects or interactions between PS × LOC and ED × LOC into SPSS v.22 for further analyses. We conducted repeated measures ANOVA to determine the impact of LOC (between-subjects factor) on BOLD response to the food image conditions after including relevant covariates (e.g. SNR, average motion during the fMRI scan, food image liking, age, sex, MVPA from the ActiGraph and pre-fMRI satiety rating). As in the behavioural analyses, covariates were tested individually to avoid overwhelming the models.

Results

Child characteristics

Table 1 shows the child characteristics by LOC status. Of the whole sample, approximately 32% reported LOC eating in the previous 3 months. In the group who reported LOC (n = 13), each participant responded ‘yes’ to at least two additional follow-up questions on the LOC assessment. The LOC episodes were characterized primarily by hunger prior to the episode (77%), continuing to eat although full (77%) and the perception that others would think they ate too much (77%). The LOC episodes occurred either after something good (70%) or bad (30%) happened. Approximately half of the LOC group reported attempts to restrict their intake prior to the episode (46%), but few reported feeling numb or zoning out (23%), although it is important to note that follow-up regarding any responses (e.g. frequency or extent of restriction) was not part of this LOC assessment.

The majority of the sample was healthy-weight and non-Hispanic Caucasian. Those who reported LOC vs. no-LOC did not significantly differ in anthropometrics, physical activity, sedentary time or age. There were also no differences in rated liking or wanting for the food images (i.e. LOC did not rate liking/wanting of large portion or high-ED food images higher than no-LOC). For parentally reported appetitive traits, children who reported LOC rated higher for desire to drink (t_{df 49} = 2.4; P < 0.05) and food fussiness (t_{df 49} = 2.9; P < 0.01) compared with no-LOC children. There were no differences between groups for satiety responsiveness, enjoyment of food, emotional overeating, food responsiveness, slowness of eating or emotional undereating (P > 0.05 for all).

Effects of meal portion size on intake curves by loss of control status

Total weight consumed

Testing LOC as a covariate in random coefficient models revealed a significant interaction between amount served and LOC status on weight of food consumed (F_{df 1, 124} = 11.8; P < 0.005). Figure 1a when the amount of all foods served at the meal was increased in portion size, children who reported LOC showed a different response pattern than children who reported no-LOC. For those who reported LOC, the mean curve had a positive linear slope of β = 0.62; 95% confidence interval (CI): 0.26, 0.98; P < 0.005 and a quadratic deceleration of β = −0.001; 95%CI: −0.001, −0.0004; P < 0.005. That is, children who reported LOC compared with no-LOC ate an average of 62% more of the additional food served, and the rate of intake was reduced by the quadratic coefficient at the larger portion sizes. Maximal mean intake for total weight consumed was reached at the 167% condition.

Figure 1.

Meal intake curves across portion size conditions (grams and kcal) by loss of control (LOC) status. Meal intake curves for total weight consumed in grams (a) and total energy consumed in kcal (b) by LOC status as weight of food served increased. (a) Total mean intake by weight (grams ± standard error of the mean) for all foods consumed from the test meals. (b) Total mean intake by energy (kcal ± standard error of the mean) for all foods consumed from the test meals. Random coefficient models, which allow individual trajectories to vary across participants, were used to compare the mean curves of total intake in response to increases in portion size served for children who reported LOC eating (LOC, n = 15) to those who reported no-LOC (non-LOC, n = 34). Intake curves of those reported LOC and non-LOC differed significantly from each other (P < 0.005).

Total energy consumed

There was a significant interaction between amount served and LOC status on energy intake (F_{df 1,} 124 = 11.7, P < 0.005). Figure 1b in those who reported LOC, the mean curve had a positive linear slope of β = 0.92; 95%CI: 0.39, 1.45; P < 0.005 and a quadratic deceleration of β = −0.001; 95%CI: −0.002, −0.0004; P < 0.005. Children who reported LOC compared with no-LOC ate an average of 92% more kcal of the additional food served, and this rate of increase was decreased by the curve’s quadratic deceleration.

Effects of meal portion size on intake from high energy density and low energy density foods by loss of control status

High energy density foods

There was a significant portion size × LOC interaction for weight consumed (F_{df 1, 99} = 9.7, P < 0.005) Fig. 2a and for energy intake (F_{df 1, 99} = 10.3, P < 0.005) Fig. 2b from high-ED foods. For those who reported LOC, the mean curve for weight of high-ED food consumed had a positive linear slope of β = 0.45; 95%CI: 0.16, 0.74; P < 0.005 and a quadratic deceleration of β = −0.001; 95%CI: −0.001, −0.0002; P < 0.005. In those who reported LOC, the mean curve for energy intake from high-ED foods had a positive linear slope of β = 0.84; 95%CI: 0.32, 1.35; P < 0.005 and a quadratic deceleration of β = −0.001; 95%CI: −0.002, −0.0005; P < 0.005. Compared with those who reported no-LOC, those who reported LOC ate an average of 45% more by weight or 84% more by energy of high-ED foods as portions were increased from reference. Peak consumption (i.e. the vertex of the curve) occurred at the 167% condition.

Figure 2.

Higher energy density (ED) intake curves across portion size conditions (grams and kcal) by loss of control (LOC) status. Intake curves for higher-ED foods consumed across portion size conditions in (a) grams and (b) kcal by LOC status. (a) Mean intake by weight (grams ± standard error of the mean) for higher-ED foods from the test meals. (b) Mean intake by energy (kcal ± standard error of the mean) for higher-ED foods from the test meals. Random coefficient models, which allow individual trajectories to vary across participants, were used to compare the mean curves of intake in response to increasing portion size served in children who reported LOC eating (LOC, n = 15) to those who reported no-LOC (non-LOC, n = 34). Mean intake curves for higher-ED foods of those reporting LOC differed significantly from non-LOC in (a) grams and (b) kcal (P < 0.005 for both).

Low energy density foods

Loss of control status did not significantly influence the PSE curve for intake by weight (P > 0.10) or energy consumed (P > 0.10) from low-ED foods.

Brain response to portion size food cues by loss of control status

Main effects of portion size and energy density

We previously reported on the main effects of portion size and ED in the first 36 children from this cohort (38), so these results will not be reported in detail here. However, because we had a larger sample size in the present study, we ran those contrasts again with the updated sample and found additional main effects of portion size in the left lingual gyrus (large portion > small portion; [F_{df 1, 48} = 37.76; P < 0.01]) and superior temporal gyrus (small portion > large portion; [F_{df 1, 48} = 15.43; P < 0.01]). For ED, the main effects overlapped with what has been previously reported, with significant differences in response to High-ED vs. Low-ED foods in the fusiform gyrus, cerebellar declive, medial frontal and superior temporal gyrus. As these effects have been previously reported (38), they will not be discussed further in the current paper.

Loss of control × portion size

There was a significant interaction between LOC status and portion size condition on BOLD response in the left cerebellum (F_{df 1, 48} = 22.84; P < 0.01). Children who reported LOC compared with no-LOC had increased activation in the left cerebellum (x, y, z = −46, −54, −39) when comparing the response to small relative to large portion sizes, regardless of the ED of the images. Tukey’s post hoc test also confirmed that children who reported LOC had greater BOLD response in the left cerebellum to small relative to large portion cues, but this difference was not present in no-LOC children (Fig. 3).

Figure 3.

Activation in the left cerebellum when viewing large relative to small portion food cues by loss of control (LOC) status. Parameter estimates (±standard error of the mean) for average activation by LOC status to large relative to small portion size food cues shown during functional magnetic resonance imaging. Children who reported LOC compared with no-LOC had significantly lower activation in the left cerebellum when viewing large relative to small portion size (PS) food cues (* indicates significantly different from one another in Tukey’s post hoc tests at P < 0.001). Functional magnetic resonance imaging statistical maps in coronal view showing activation to large relative to small portion size food cues and co-registered with average structural magnetic resonance imaging data from participants. Results are from whole-brain analyses (n = 48) that tested the effects of portion size by LOC status in representative F maps showing increased blood oxygenation level dependent activation (red colours) to small relative to large food portion sizes in the left cerebellum (P < 0.001, corrected). l, left.

Loss of control × energy density

There was a significant interaction between LOC status and ED condition on BOLD response in the right cerebellum (F_{df 1, 48} = 21.05; P < 0.01). Children who reported LOC compared with no-LOC showed significantly increased activation in the right cerebellum (x, y, z = 6, −52, −39, −28), when viewing all High-ED relative to Low-ED food cues, regardless of their portion size. Tukey’s post hoc test also showed that for High-ED food cues, children who reported LOC engaged the right cerebellum to a greater extent than children with no-LOC, while for Low-ED foods, the opposite response pattern was observed (Fig. 4).

Figure 4.

Activation in the right cerebellum when viewing High energy density (ED) vs. Low-ED food cues by loss of control (LOC) status. Parameter estimates (±standard error of the mean) for average activation by LOC status to High-ED relative to Low-ED food cues shown during functional magnetic resonance imaging. Children who reported LOC compared with no-LOC had significantly higher activation in the right cerebellum when viewing High-ED relative to Low-ED food cues. (* indicates significantly different from one another in Tukey’s post hoc tests at P < 0.001). Functional magnetic resonance imaging statistical maps in saggital view showing activation to High-ED relative to Low-ED food cues and co-registered with average structural magnetic resonance imaging data from participants. Results are from whole-brain analyses (n = 48) that tested the effects of ED by LOC status in representative F maps showing increased blood oxygenation level dependent activation (red colours) to high-ED relative to Low-ED food cues in the right cerebellum (P < 0.001, corrected). PS, portion size; r, right.

Discussion

These findings provide new evidence that children who report LOC may be vulnerable to the effects of increasing portion size at a meal, particularly for high-ED foods. The pattern of intake differed across portion size conditions in children who reported LOC compared with no-LOC. Notably, these findings were robust and independent of both child weight status and how much children liked the foods. Previous studies have linked LOC to obesity (20) and future weight gain (26). Furthermore, children in the current study who reported LOC compared with no-LOC showed greater activation in response to food cues to small vs. large portion and High-ED vs. Low-ED food cues in the cerebellum, a prominent area of the hindbrain implicated in emotional processing (55) and satiety signalling (56). The cerebellum is also highly connected to the insular cortex, an important appetitive processing region (57). Our results suggest that children who report LOC in middle childhood may be susceptible to overconsumption from large portions of high-ED foods and have altered processing of these food cues in the brain.

Children who reported LOC compared with no-LOC had different PSE curves for intake, specifically from high-ED foods. Results were not changed when adjusting analyses for liking of high-ED foods served at the meals, suggesting that these findings were not driven by subjective ratings of palatability alone. In contrast, LOC status did not impact the PSE curve for low-ED foods. Our results are supported by past work showing that children with LOC compared with those without have different energy intake patterns (30) and may have selective bias towards high-ED foods (29,58). Eating-related LOC is linked to disinhibited eating patterns such as emotional eating and negative affective states (59–61), indicating that those who report LOC may turn to higher ED foods as a coping strategy to improve mood. In addition, children who report LOC also tend to have more impulsivity and attention deficits (62). Therefore, LOC reporters may respond more impulsively to higher ED foods, particularly when hungry. Another possibility is that those with LOC have impaired ability to regulate their intake of palatable, more energy-dense foods in larger portions than those without LOC, but additional studies are needed to confirm this.

Additional insight into the behavioural findings was provided through our fMRI results, which show differences in cerebellar food-cue processing between those who reported LOC compared with those who did not. A recent study showed that girls with LOC show differential brain response in the ventromedial prefrontal cortex and striatum when under social distress, suggesting that dysregulated brain response may be linked to subsequent overeating (35). Although there are important differences in design between that study (35) and the present, together the findings suggest that children who report LOC compared with no-LOC may have altered brain processing of food cues and that the cerebellum may be one of the multiple important neural targets for future studies.

The whole-brain analyses presented in this study found that children who reported LOC compared with no-LOC had (i) increased activation when viewing High-ED relative to Low-ED food cues in the right cerebellum and (ii) decreased activation when viewing large relative to small portion size food cues in the left cerebellum. The cerebellum is functionally diverse with substantial projections to non-motor areas involved in drug craving (63), reward processing (64), affective processing (55) and emotional regulation (65). Furthermore, closed-loop circuits in the cerebellum involving the thalamus, cortical areas (66) and deep cerebellar nuclei are interconnected with the hypothalamus (67). Cerebellar activation is commonly reported in fMRI tasks (generally) and in relation to food cues (specifically) in adults (47,68), adolescents (69) and children (32). In particular, the cerebellum may assist in directing satiety signals to other brain regions for higher-level processing (56). Our observed responses to High-ED food cues in LOC reporters align with those of an fMRI study finding activation only in the cerebellum to energy-dense food cues in healthy-weight, restrained eaters (70). Restrained eaters cognitively restrict intake as an attempt to control body weight but are also more vulnerable to counter-regulatory overeating (e.g. binge-eating) (71). It is possible that High-ED food cues prime the cerebellum, which is densely populated with leptin receptors (72) and functionally connected to the insula to form a ‘homeostatic-circuit’ (73). High-ED foods also tend to provide less satiation, which may delay feelings of fullness for those who are prone to LOC while eating. Unexpectedly, though, we saw that large portions engaged the left cerebellum to a lesser extent than small portions. These findings are somewhat inconsistent with observations that the LOC reporters had greater BOLD response in the right cerebellum to higher ED food cues. One speculation is that children who exhibit LOC eating may be processing the amount of food presented incorrectly, which may impair the development of satiation cues at a meal. If these findings are replicated, future studies may wish to investigate strategies such as cognitive reappraisal (74) of larger portion, High-ED food cues to help LOC reporters moderate excess consumption.

One strength of this study includes the comprehensive design for detecting individual differences in the response to portion size across multi-item meals. Previous studies, although well-designed, have used fewer than four portion size conditions (1,8,9). Incrementally, increasing the portion size served of all foods across four counterbalanced conditions allowed the comparison of intake curves by LOC status (11). Because previous studies have not observed clear differences in portion size response by weight status (8,10), future studies examining children during middle childhood may also wish to incorporate LOC to understand variability in the PSE. Another strength is that we conducted fMRI and voxel-wise analyses to shed light on the underlying neurocircuitry implicated in how children process food cues varied by portion size and ED. While useful for generating hypotheses, additional studies are needed to characterize the neural connections between regions associated with LOC status. Although the inclusion of fMRI allowed investigation of brain responses to food cues, these responses may not generalize to those experienced during actual eating events. Furthermore, we cannot extrapolate these laboratory findings to eating behaviours outside the laboratory or to different foods. Our measure of LOC incidence was dichotomized by its presence or absence, and therefore, future studies may want to characterize children by severity or variety of LOC symptoms. As such, we recognize the limited generalizability of LOC as characterized in this study in that it may not be generalizable to LOC present in clinically diagnosed eating disorders. In addition, LOC was reported by more boys than girls, which contradicts one previous report in the literature (22), although this may align with national data that suggest male-reporting of binge-eating is more common than other disordered eating (75). Importantly, the discrimination of LOC in youth by sex, among other factors, is inconsistent across the literature (12) and needs further exploration.

Conclusion

Results from this study show that eating-related LOC has a robust influence on children’s susceptibility to increased portion size at a multi-item meal. Further, these are the first results to indicate that brain responsiveness to food portion size and ED cues varies by LOC status. There are implications of this work for obesity prevention. Limiting larger portions of high-ED foods, while important for all children, may be an especially useful strategy to moderate excess consumption in children who experience LOC.