Does ‘portion size’ matter? Brain responses to food and non-food cues presented in varying amounts

Abstract

Larger portions of food elicit greater intake than smaller portions of food, particularly when foods are high in energy density (kcal/g; ED). The neural mechanisms underlying this effect remain unclear. The present study used fMRI to assess brain activation to food (higher-ED, lower-ED) and non-food (office supplies) images presented in larger and smaller (i.e., age-appropriate) amounts in 61, 7–8-year-olds (29 male, 32 female) without obesity. Larger amounts of food increased activation in bilateral visual and right parahippocampal areas compared to smaller amounts; greater activation to food amount (larger>smaller) in this cluster was associated with smaller increases in food intake as portions increased. Activation to amount (larger > smaller) was stronger for food than office supplies in primary and secondary visual areas, but, for office supplies only, extended into bilateral parahippocampus, inferior parietal cortex, and additional visual areas (e.g., V7). Activation was greater for higher- vs. lower-ED food images in ventromedial prefrontal cortex for both larger and smaller amounts of food; however, this activation extended into left lateral orbital frontal cortex for smaller amounts only. Activation to food cues did not differ by familial risk for obesity. These results highlight potentially distinct neural pathways for encoding food energy content and quantity.

1. Introduction

Larger portions of food, particularly those high in energy density (ED; kcal/g), elicit greater intake than smaller portions, termed “the portion size effect” (PSE). This effect has been demonstrated across food type (Reale et al., 2019), environmental context (Diliberti et al., 2004; Gough et al., 2021), and developmental stage (Reale et al., 2019), and is sustained over 5 days in children (Smethers et al., 2019). Nevertheless, the mechanisms underlying this effect remain elusive (English et al., 2015; Herman et al., 2015). Large portions of high-ED food may drive excess food intake by influencing reward (Burger et al., 2011), cognitive (Marchiori et al., 2014), and/or perceptual processes (McCrickerd and Forde, 2016). Examining brain responses to visual food cues that vary by amount (i.e., portion size) and ED can shed light on how these food characteristics interact to influence eating behaviors.

Previous studies using functional magnetic resonance imaging (fMRI) characterized brain responses to food cues in general as well as food ED, but brain responses to food amount (i.e., portion size) are less clear. Meta-analyses show that cues depicting palatable, typically higher-ED foods activate brain regions associated with reward and valuation, response inhibition, and visual processing more than non-food related cues (van der Laan et al., 2011; van Meer et al., 2015). Brain regions implicated in reward, cognitive, and visual processes also show greater activation to high- compared to low-energy foods (van der Laan et al., 2011). While fewer studies have characterized the brain response to food images that vary in portion size, our lab previously showed children aged 7–10 years had greater activation in regions implicated in inhibitory control (inferior/orbital frontal gyri; English et al., 2017) and visual processing (left lingual gyrus; English et al., 2019) when viewing images of larger than smaller food portions. However, these prior studies presented only food items in varying amounts; thus, it was not possible to examine whether brain responses to amount differ between food and non-biologically relevant stimuli. While differences in image properties (e.g., object size, complexity) can influence perceptual processing (e.g., in early visual cortex; Eger et al., 2008), food is a primary reinforcer. Therefore, variations in the amount of food depicted may differentially engage regions sensitive to reward magnitude (e.g., medial orbitofrontal cortex, striatum; (Diekhof et al., 2012). Comparing brain responses to amount (i.e., larger vs. smaller) between food and non-food items will help identify brain regions uniquely sensitive to food amount. Additionally, as higher-ED foods differentially influence brain processes (van der Laan et al., 2011) and increase energy intake (Kral et al., 2004) relative to lower-ED foods, brain response to food amount may also differ by ED (English et al., 2017). Assessing the interaction between food amount and ED will shed light on how large portion sizes specifically impact brain responses to food items that drive overconsumption.

In addition to identifying brain regions that are responsive to visual depictions of food amount, the present study also advances the literature by examining food cue reactivity in a cohort of children without obesity who vary by familial risk for obesity. Examining food cue reactivity in children without obesity may reduce variability in brain responses due to body weight status (Bruce et al., 2010; van Meer et al., 2019), increasing statistical power to detect effects of food characteristics on brain activity. Second, stratifying children by familial risk for obesity can provide insight into neurobiological traits that may elevate risk for excess energy intake and weight gain. Children at higher familial risk for obesity are twice as likely to develop obesity themselves (Wang et al., 2017) and have reduced prefrontal cortical and anterior cingulate activation to food vs. non-food images (Luo et al., 2021) and high- vs. low-ED food words (Carnell et al., 2017). However, it remains unclear how familial risk impacts brain responses to images that vary in ED and portion size.

To better understand the influence of portion size on brain food cue reactivity, this study used fMRI to examine brain responses to food and non-food images presented in varying amounts. To build on prior work from our lab (English et al., 2019, 2017), analyses (1) were conducted in a sample that was larger, more homogeneous in age and weight status, and characterized by familial obesity risk and (2) included explicit comparisons between responses to amount for food and non-food cues. We hypothesized that (1) larger amounts of food would elicit greater activation than smaller amounts of food in brain regions associated with reward, cognitive control, and visual processing and that (2) activation to food amount (larger > smaller) would be greater than activation to non-food amount (larger > smaller) in brain regions associated with reward, cognitive control, and visual processing. To gain insight into risk for childhood obesity, we compared activation to food characteristics (i.e., energy density and amount) between children at high and low familial risk for obesity. We hypothesized that, compared to children at low familial risk, children with high familial risk would show (1) greater activation to energy density (higher > lower) and food amount (larger > smaller) in brain regions associated with reward processing and (2) decreased activation to energy density (higher > lower) and food amount (larger> smaller) in brain regions associated with cognitive control.

2. Methods

2.1. Participants

Participants in this study were 88 (45 male, 43 female) 7–8-year-old children (mean [SD] age=7.8 [0.62] years) who attended the MRI session (visit 6) of a 7-visit prospective study. Children were accompanied by a parent from each family (76 mothers, 12 fathers). Children were without obesity (BMI-for-age-and-sex percentile < 90) by study design and were classified as having high (n=36; maternal BMI > 30 kg/m² ± 1 unit) or low (n=52; maternal BMI < 25 kg/m² ± 1 unit) familial risk for obesity (Keller et al., 2023). Exclusion criteria included the biological mother not meeting BMI requirements for one of the familial risk groups and parents reporting their child was colorblind, not reading at grade level, not fluent in English, had a learning (e.g., dyslexia) or neurodevelopmental (e.g., ADHD) disability, had a diagnosed psychological condition (e.g., anxiety), was taking medications known to influence appetite, cognition, or blood flow, or had any MRI contraindications (e.g., metal in the body, claustrophobic). 61 children were included in analyses while 27 were excluded (see 3.3.3). Children excluded from analyses did not differ demographically from those who were included (see 4.1).

2.2. Design

As part of a prospective study aiming to identify neurocognitive risk factors for pre-adolescent obesity (ClinicalTrials.gov NCT03341247), participants attended six baseline visits and one follow-up visit a year later (study protocol and materials available on Open Science Foundation (OSF) at https://osf.io/ynjqw/). Primary assessments included child body composition, actigraphy, laboratory eating behaviors, and brain food cue reactivity. Methods and results related to meal intake, cognitive and behavioral assessments, and resting-state functional connectivity have been published elsewhere (Gunther et al., 2022; Keller et al., 2023; Zuraikat et al., 2023). Visits took place at the University Park campus of The Pennsylvania State University. Parental consent and child assent were obtained in accordance with the Institutional Review Board of The Pennsylvania State University. Baseline data collection occurred between 2017 and 2022 but was interrupted for ~8 months due to COVID-19.

Data for the present analyses included parent-reported demographics (visit 1; see section 2.3), child and parent anthropometrics (visit 1; see section 2.4), portion size meals (visits 2–5; see section 2.5), and brain food cue reactivity (visit 6; see sections 2.6–2.8). Children completed a mock-MRI protocol on visits 4 and 5 (see supplement). Children fasted for at least 3 hours prior to each visit and fullness was measured upon arrival using an age-appropriate visual analog scale (Keller et al., 2006).

2.3. Demographics

The accompanying parent reported child race, child ethnicity, yearly family income and parental education in a demographic questionnaire (see Table 1). Family income and maternal education were used as proxies for socioeconomic status (Bradley and Corwyn, 2002).

Table 1.

Descriptive statistics for demographic and MRI-related variables by familial risk for obesity

	Low Risk	High Risk	Overall
	(N=36)	(N=25)	(N=61)
Age, yrs
Mean (SD)	7.9 (0.63)	7.7 (0.55)	7.8 (0.60)
Min, Max	7.02, 8.99	7.00, 8.84	7.00, 8.99
BMI percentile
Mean (SD)	41.5 (25.7)	52.8 (23.6)	46.2 (25.3)
Min, Max	3.91, 86.8	9.38, 89.0	3.91, 89.0
Fat mass index
Mean (SD)	4.10 (0.80)	4.81 (0.86)	4.39 (0.89)
Min, Max	3.18, 6.44	3.57, 7.42	3.18, 7.42
Sex; n (%)
Male	18 (50.0%)	11 (44.0%)	29 (47.5%)
Female	18 (50.0%)	14 (56.0%)	32 (52.5%)
Race; n (%)
White	34 (94.4%)	25 (100%)	59 (96.7%)
American Indian/Alaskan Native	0 (0%)	0 (0%)	0 (0%)
Asian	2 (5.6%)	0 (0%)	2 (3.3%)
Black/African American	0 (0%)	0 (0%)	0 (0%)
Hawaiian/Pacific Islander	0 (0%)	0 (0%)	0 (0%)
Ethnicity; n (%)
Not Hispanic or Latino	36 (100%)	25 (100%)	61 (100%)
Hispanic or Latino	0 (0%)	0 (0%)	0 (0%)
Family Income; n (%)
< $51,000K	4 (11.1%)	5 (20.0%)	9 (14.8%)
51–100K	13 (36.1%)	14 (56.0%)	27 (44.3%)
>$100K	18 (50.0%)	5 (20.0%)	23 (37.7%)
Missing	1 (2.8%)	1 (4.0%)	2 (3.3%)
Maternal Education; n (%)
< BA	3 (8.3%)	10 (40.0%)	13 (21.3%)
BA	16 (44.4%)	13 (52.0%)	29 (47.5%)
> BA	16 (44.4%)	2 (8.0%)	18 (29.5%)
Missing	1 (2.8%)	0 (0%)	1 (1.6%)
Analyzed runs, n; Mean (SD)	4.7 (0.58)	4.2 (0.99)	4.5 (0.81)
Analyzed timepoints, n; Mean (SD)	350 (52)	310 (82)	330 (69)
Average framewise displacement; Mean (SD)	0.32 (0.19)	0.45 (0.29)	0.37 (0.24)
Pre-MRI fullness; Mean (SD)	73 (30)	67 (28)a	71 (29)1
Pre-MRI anxiety; Mean (SD)	1.8 (1.6)	2.7 (3.1)	2.2 (2.3)
Consumed snack; n (%)
no	13 (36.1%)	8 (32.0%)	21 (34.4%)
yes	23 (63.9%)	16 (64.0%)	39 (63.9%)
Missing	0 (0%)	1 (4.0%)	1 (1.6%)

Note: table reflects data for children included in analyses; BA: bachelor’s degree

^a1 value missing. Value was imputed for neuroimaging analyses but not included in descriptive statistics.

2.4. Anthropometrics

Height and weight of children and their accompanying parents were measured twice by a trained researcher using a standard scale (Scale Tronix model 5002, Welch Allyn, Chicago, IL; precision to nearest 0.1 kg) and stadiometer (precision to nearest 0.1cm) with shoes and heavy clothing items removed. Height and weight for the parent who did not attend the visit were reported by the accompanying parent. Maternal BMI was calculated from average measured (86%) or parentally reported (14%) maternal height and weight. Child BMI was calculated from averaged height and weight with BMI-for-age-and-sex percentile computed using the Centers for Disease control growth charts (Kuczmarski et al., 2002). Children also completed whole-body dual energy x-ray absorptiometry (DEXA; Hologic Inc., Waltham, MA), which was used to calculate fat mass index (fat mass (kg)/height, m²).

2.5. Portion size meals

Details about the meals that were varied in portion size were described previously (Keller et al., 2023). Briefly, children were served 4 meals of varying portion sizes in a counterbalanced order. All meals were served in the laboratory at the child’s usual meal times. Portion size conditions included a reference amount and 133%, 166%, and 199% of the reference condition by food weight. Food items (i.e., chicken nuggets, macaroni and cheese, grapes, and steamed broccoli) were consistent across meals and access to water was provided ad libitum. Children were given 30 minutes to eat ad libitum until they reached satiation and could notify the researcher if they reached satiation before 30 minutes. Food weight was assessed before and after each meal to the nearest 0.1 g. Prior to each meal, children reported liking in response to tasting small samples (~2–3g) of the meal foods using a five-point facial hedonic scale and reported fullness using a child-friendly visual analog scale (Keller et al., 2006).

2.6. Visit 6 procedures

To scan children in a neutral appetitive state, those who reported fullness levels <25% upon arrival received a small snack (6.75 fl oz apple juice, Quaker Chewy granola bar). Following consumption, children rated their fullness again. If fullness rating remained <25%, a second snack was offered. Children also rated their state anxiety before and after the scan using the Children’s Anxiety Meter Scale (CAMS; Ersig et al., 2013). Following the scan, children viewed and rated each image of the food-cue task for liking and anticipated fullness using a computerized visual analog scale. For each image, children were instructed to answer, “how much do you like this object/food?” (0 = not at all; 100 = like very much). For each food image, children were also instructed to answer, “how full would your stomach be if you ate this food?” (0 = not at all full; 100 = very much full). Children rated images in the same order they were presented in the scanner.

2.7. fMRI acquisition

fMRI was conducted using a Siemens MAGNETOM Trio 3 T MRI scanner with a 20-channel head coil. Padding was placed around the head to limit movement. Stimuli were projected onto an MRI-compatible screen and viewed by the child using a mirror. A T1-weighted MPRAGE was acquired using Siemens’ integrated parallel acquisition technique (iPAT) generalized autocalibrating partially parallel sequences (GRAPPA, acceleration factor 2; Griswold et al., 2002) and the following parameters: TR=1650, TE=2.03 ms, flip angle=9°, FOV=256×256 mm², 160 sagittal slices, voxel size 1×1×1 mm³. Functional scans were performed to assess the blood oxygen dependent response (BOLD), an indirect measure of neuronal activation (Arthurs and Boniface, 2002). Functional scans were collected using a T2*-weighted gradient single-shot echo planar imaging sequence (TR=2000 ms, TE=26 ms, flip angle=90°, FOV=220×220 mm²) with a voxel size of 3×3×3 mm³ to acquire 33 axial slices (no gap) in descending order. Functional images were aligned along the AC-PC plane and adjusted vertically to optimize signal in the temporal lobe and cerebellum based on findings in these regions from our group (English et al., 2017). Six functional runs were collected including one resting-state scan (180 volumes; Gunther et al., 2022) and five food-cue scans (80 volumes each). Two volumes were automatically discarded prior to the start of each functional run. Following the functional runs, a field map was acquired using a double-echo gradient-echo sequence (TR=400 ms, TE₁=5.12 ms, TE₂=7.65 ms, flip angle=60°, FOV=220×220 mm²) with a voxel size of 3×3×3 mm³ to acquire 33 axial slices (no gap) in descending order. MRI data are available on OpenNeuro (Fuchs et al., 2023).

2.8. fMRI food-cue task

The food-cue task was administered using E-Prime 2.0 (Psychology Software Tools, Inc., n.d.). Children viewed 120 food and 60 non-food (i.e., office supply) images from a standardized dataset (Kling et al., 2020). Prior to the task, children were instructed to answer the question “Do you want this?” for each item by selecting either “Do not want/Frown face” or “Want/Smiley face” with their dominant hand. Office supplies were selected for the non-food condition because they are not biologically salient, but they are familiar to children and visual characteristics (e.g., size, color, contrast) could be balanced with food items. Food conditions varied by amount (larger, smaller) and ED (higher, lower), while office supply conditions varied by amount (larger, smaller) only. Higher-ED foods (e.g., bacon, brownie) had >2.0 kcal/g, while lower-ED foods (e.g., strawberries, turkey) had <1.5 kcal/g. The portions of food depicted in the smaller amount condition were based on standard servings from the Nutrition Facts Panel and reflected the amount of food typically consumed by children (Smiciklas-Wright et al., 2003). The portions of food depicted in the larger amount condition were approximately double the weight of food shown in the smaller amount condition. Images were presented over five runs of six blocks each (one per condition; Figure 1).

Figure 1.

Food-cue task. Children viewed 120 food and 60 non-food (i.e., office supply) images from a standardized dataset (Kling et al., 2020). Food conditions varied by amount (larger, smaller) and energy density (ED; higher, lower), while office supply conditions varied by amount (larger, smaller) only. For each image, children indicated whether they wanted the item by pressing one of two buttons (thumb for “yes”, index finger for “no”) on a response grip with their dominant hand. (A) Image presentation: Images were presented on a black screen above a smiley face and a frowny face, which were included to remind children to make a response. (B) Example run: Images were presented over five runs. Each run contained six blocks (one per image condition) with six images per block. Images were presented for 2.0 s with a 0.5 s inter-trial fixation cross. Each block was followed by an 8.0 s white fixation cross presented on a black screen. Two versions of the task with pseudorandomized block orders were counterbalanced across participants.

2.9. Analyses

Code is available OSF (https://osf.io/x8j56/).

2.10. Descriptive statistics

We compared demographic variables between children at high and low familial risk for obesity using 2-sample t-tests for continuous variables and χ2 or Fisher’s Exact Test for categorical variables. To ensure there were no meaningful differences between children included and excluded (e.g., due to motion) from analyses, we examined demographic differences by inclusion status.

We tested associations between MRI-related variables (e.g., pre-MRI fullness, average framewise displacement—FD) and familial risk for obesity, age, and sex using Pearson’s correlation, 2-sample t-tests, and χ². Pre-MRI fullness was indexed as the fullness rating that occurred closest to the scan with one value imputed due to recording error (see supplement).

2.11. fMRI preprocessing

MRI data were preprocessed using fMRIPrep 20.2.3 (Esteban et al., 2017) using standard pre-processing steps (see supplement for full report). Volume-based spatial normalization used Montreal Neurological Institute’s (MNI) unbiased template for pediatric cohort 3 (Fonov et al., 2011, ages 7–11y; 2009). Functional preprocessing included susceptibility distortion correction, estimation of head-motion parameters, slice-time correction, co-registration to anatomical space, and resampling into standard space. Quality assessment reports generated by fMRIPrep were visually inspected, and data were excluded (see 3.3.3) or re-processed as needed. Subsequent processing and individual- and group-level analyses were conducted using AFNI v21.3.04 (Cox, 1996; Cox and Hyde, 1997); this included applying a 6.0 mm FWHM BOLD Gaussian blur and scaling the time-series data so results can be represented as percent signal change.

2.12. Percent wanting

For each food-cue task block, we calculated the percent of items children reported wanting out of items responded to (%want). We used two separate linear mixed-effects models using lme4 (Bates et al., 2015) in R 4.2.2 (R Core Team, 2021) with subject as random effect to assess (1) the interaction between cue type (food, office) and amount (larger, smaller) on %want and (2) the interaction between ED (higher-ED, lower-ED) and food amount (larger, smaller) on %want, controlling for run.

2.13. Image ratings

Similar to analyses of %want, we calculated children’s average liking and anticipated fullness ratings for each food-cue task block. Two separate linear mixed-effects models with subject as random effect were used to assess the interactions between (1) cue type (food, office) and amount (larger, smaller) and (2) ED (higher-ED, lower-ED) and food amount (larger, smaller) on liking, controlling for run. In addition, a linear mixed-effects model with subject as random effect was used to assess the interaction between ED (higher-ED, lower-ED) and food amount (larger, smaller) on anticipated fullness, controlling for run.

2.14. fMRI Analyses

2.14.1. Individual-level

Using AFNI’s 3dDeconvolve, we conducted individual-level general linear models (GLMs) that included 14 nuisance regressors computed by fMRIPrep (6 rigid-body motion parameters and their derivatives, average signal within the cerebrospinal fluid and white matter masks). Each image condition (i.e., larger amount higher-ED foods, larger amount lower-ED foods, larger amount office supplies, smaller amount higher-ED foods, smaller amount lower-ED foods, smaller amount office supplies) was modeled by convolving block onsets and durations to the hemodynamic response (AFNI’s ‘BLOCK’ function). From these GLMs, we generated a total of 8 first-level contrasts: (1) food > office, (2) higher-ED foods (larger + smaller amounts) > lower-ED foods (larger + smaller amounts), (3) larger amount higher-ED foods > larger amount lower-ED foods, (4) smaller amount higher-ED foods > smaller amount lower-ED foods, (5) larger food amounts (higher-ED + lower-ED) > smaller food amounts (higher-ED + lower-ED), (6) larger amount higher-ED foods > smaller amount higher-ED foods, (7) larger amount lower-ED foods > smaller amount lower-ED foods, and (8) larger amount office supplies > smaller amount office supplies. We also conducted separate GLMs to assess the parametric modulation of BOLD responses to office supplies, higher-ED foods, lower-ED foods, larger food amounts, and smaller food amounts by %want.

To remove signal outliers and high-motion timepoints from analyses, volumes were censored by 3dDeconvolve if they (1) were identified as steady state outliers by fMRIPrep or (2) had a FD>0.9mm (Power et al., 2014), respectively. The FD threshold selected matches that used by the Adolescent Brain Cognitive Development (ABCD ^®) Study (Hagler et al., 2019). Runs were excluded from first-level analyses if >20% of volumes across food- and office supply blocks were censored.

2.14.2. fMRI exclusion criteria

We excluded children from fMRI analyses if they declined to scan (n=4), completed <3 runs of the food-cue task (n=4), had <3 runs of the food-cue task included in first-level analyses after censoring (see 3.2.2; n=17), or had poor-quality functional data (e.g., extreme loss of field of view due to motion) detected via visual inspection (n=2). Children were also excluded from parametric analyses if they had 1 or more food-cue blocks with no response data (n=4 excluded from all models) or no variability in %want for 1 or more of the conditions included in the model (n excluded: office supply=8; ED=3; food amount=2).

2.14.3. Group-level

Group-level analyses were conducted using 3dttest++. Z-statistic maps were thresholded at voxel-wise p<0.001 (two-tailed) and cluster-corrected to p<0.05 except for exploratory analyses with familial risk status and fat mass index, which were cluster-corrected to p<0.1. 3dClustStim was used to determine cluster extents (Cox et al., 2017; Table S2). Unthresholded maps are available on Neurovault.org (https://identifiers.org/neurovault.collection:14535; Gorgolewski et al., 2015). Models included average FD, sex, pre-MRI fullness, and pre-MRI CAMS as covariates and were restricted to voxels scanned in at least 80% of subjects included in the given analysis (see supplement; Figure S1). We conducted sensitivity analyses by testing models with snack intake (yes/no) as an additional covariate; results were similar with the addition of this covariate (Tables S5–S10), so results from models without the snack intake as a covariate are reported in the main text.

2.14.3.1. BOLD responses to image condition

To assess the main effects of cue type (food vs. office), ED (higher vs. lower), and food amount (larger vs. smaller) on BOLD responses, we conducted three 1-sample t-tests on the following first-level contrasts: (1) food > office, (2) higher-ED foods (larger + smaller amounts) > lower-ED foods (larger + smaller amounts), and (3) larger food amounts (higher-ED + lower-ED) > smaller food amounts (higher-ED + lower-ED).

To visualize similarities (i.e., conjunction) and differences (i.e., disjunction) in the extent of BOLD responses to ED for larger and smaller amounts of food, we conducted and overlayed binarized results of two 1-sample t-tests on the following first-level contrasts: (1) larger amount higher-ED foods > larger amount lower-ED foods and (2) smaller amount higher-ED foods > smaller amount lower-ED foods.

Similarly, to visualize similarities and differences in the extent of BOLD responses to amount for higher-ED foods, lower-ED foods, and office-supplies, we conducted and overlayed binarized results of three 1-sample t-tests on the following first-level contrasts: (1) larger amount higher-ED foods > smaller amount higher-ED foods, (2) larger amount lower-ED foods > smaller amount lower-ED foods, and (3) larger amount office supplies > smaller amount office supplies.

To formally test whether BOLD responses to amount differed by cue type or ED, we conducted two paired t-tests on first-level amount (larger > smaller) contrasts.

2.14.3.2. Associations with %want

To assess how in-scanner wanting related to BOLD responses, we conducted five 1-sample t-tests on the parametric modulation of BOLD responses by %want for (1) office supply, (2) higher-ED food, (3) lower-ED food, (4) larger food amount, and (5) smaller food amount conditions. We conducted two paired t-tests to assess whether the parametric modulation of BOLD responses by %want differed by (1) ED (higher vs. lower) or (2) food amount (larger vs. smaller).

2.14.3.3. Associations with obesity risk and fat mass index

To explore whether familial risk for obesity or fat mass index were associated with BOLD responses, we conducted six regressions with risk status (high vs. low) and fat mass index as separate predictors of the following first-level contrasts: (1) food > office, (2) higher-ED foods (larger + smaller amounts) > lower-ED foods (larger + smaller amounts), and (3) larger food amounts (higher-ED + lower-ED) > smaller food amounts (higher-ED + lower-ED). Associations with risk status were assessed explored using the whole analysis sample as well as only in children with measured maternal height and weight (N = 52).

2.14.3.4. Post-hoc associations with the PSE

To assess whether BOLD responses in clusters responsive to image conditions were associated with eating behaviors, we extracted first-level average contrast values from clusters exhibiting significant effects of cue type, ED (across amount), food amount (across ED), and office supply amount, as well as differential responses to amount by cue type (analyses described in section 2.14.3.1). The contrast values extracted corresponded to the contrast that the cluster was significantly responsive to (i.e., for a cluster responsive to cue type, we extracted the average BOLD response to cue type across all voxels in that cluster). Using extracted contrast values, we conducted 2 linear mixed-effects models for each cluster using lme4 (Bates et al., 2015) in R 4.2.2 (R Core Team, 2021). Models included subject as a random effect and predicted either the weight (grams) or energy (kcal) of food consumed from (1) the extracted BOLD response, (2) the proportion increase in meal size from the reference portion (i.e., meal size condition, [0.00, 0.33, 0.66, 0.99]) and (3) the interaction between the extracted BOLD response and meal size condition, controlling for pre-meal fullness, average food liking, meal order, sex and BMI.

3. Results

3.1. Participant characteristics (Table 1)

Most parents reported their children were White, non-Hispanic/Latinx, the child’s mother had a bachelor’s degree or higher, and the yearly family income was greater than $51,000. Children at low familial risk for obesity were more likely to have mothers who completed a bachelor’s degree (Fisher’s exact p<0.001) and had lower fat mass indices (p<0.01) than those at high familial risk. Age, sex, BMI percentile, and income did not differ by familial risk for obesity (ps>0.05). Children included in analyses did not differ from those excluded by age, sex, BMI percentile, fat mass index, income, maternal education, or familial risk for obesity (ps>0.43; Table S1).

On average, children included in analyses had acceptable data for 4.48 out of 5 possible runs and 332 out of 398 possible time points. Children at low familial risk for obesity had acceptable data for more runs and timepoints (ps<0.05) and lower average FD (p=0.05) than children at high familial risk for obesity. Pre-MRI anxiety, fullness, and snack consumption did not differ by risk (ps>0.17). MRI-related variables were not associated with age or sex (p>0.08) except for snack consumption, which was more frequent in girls (80%) than boys (50%) (χ²(1)=5.55, p=0.02).

3.2. Percent wanting

%want was higher for foods (estimated marginal mean (EMM)=67.5%, SE=2.4%) compared to office supplies (EMM=41.8%, SE=2.5%; p<0.001, Figure 2A). There was no main effect of amount or interaction between cue type and amount on %want (ps>0.42). Similarly, %want was greater for higher-ED foods (EMM=78.2%, SE=2.4%) than lower-ED foods (EMM=57.0%, SE=2.4%; ps<0.001; Figure 2B). There was no main effect of food amount or interaction between ED and food amount on %want (ps>0.31).

Figure 2.

Adjusted percent wanting (%want) of items by cue type (left; food vs. office supplies) and energy density (right; higher vs. lower) presented in larger (dark grey) and smaller (light grey) amounts. Percent wanting was adjusted based on linear mixed effects models with subject as random effect, controlling for run. Significance code: ***, p < 0.001.

3.3. Image liking and anticipated fullness ratings

Liking was higher for foods (EMM=58.9, SE=2.1) compared to office supplies (EMM=47.8, SE=2.2; p<0.001, Figure 3A). There was no main effect of amount or interaction between cue type and amount on liking (ps>0.49). Similarly, liking was greater for higher-ED foods (EMM=67.9, SE=2.1) compared to lower-ED foods (EMM=49.9, SE=2.1; p<0.001; Figure 6B). There was no main effect of food amount or interaction between ED and food amount on liking (ps>0.23).

Figure 3.

Image ratings by condition. Ratings were adjusted based on linear mixed effects models with subject as random effect, controlling for run. (A) Adjusted liking of items by cue type (food vs. office supplies) presented in larger (dark grey) and smaller (light grey) amounts. (B) Adjusted liking of items by energy density (ED; higher-ED vs. lower-ED) presented in larger (dark grey) and smaller (light grey) amounts. (C) Adjusted anticipated fullness by ED (higher-ED vs. lower-ED) presented in larger (dark grey) and smaller (light grey) amounts. Significance code: ***, p < 0.001.

Figure 6.

BOLD responses to amount. Coordinates are in MNI space. Z-statistic maps were thresholded at voxel-wise p < 0.001 and cluster-corrected to p < 0.05. (A) BOLD response to food amount: thresholded z-statistic map for 1-sample t-test on food amount contrast across higher- and lower-ED food conditions (i.e., larger food amounts (higher-ED + lower-ED) > smaller food amounts (higher-ED + lower-ED)). Orange clusters show regions with greater responses to larger food amounts than smaller food amounts. Blue clusters show regions with greater responses to smaller food amounts than larger food amounts. Larger and smaller amount food images are examples of photographs presented to children during fMRI. (B) Amount combination map: z-statistics maps from separate 1-sample t-tests on amount contrasts (larger > smaller) for higher-ED, lower-ED, and office supply conditions were thresholded and binarized and then combined to show conjunction and disjunction in the extent of responses to amount. Colors reflect whether voxels exhibited responses to amount (larger vs. smaller) for office supplies (blue), office supplies and higher-ED foods (pink), or all conditions (purple), or other combinations of conditions (e.g., higher-ED foods, higher-ED + lower-ED foods; yellow). (C) BOLD response to amount by cue type: thresholded z-statistic map for paired t-test comparing BOLD responses to amount (larger > smaller) between food and office supply conditions. Orange clusters show regions with greater responses to food amount than office supply amount. No regions showed greater responses to office supply amount than food amount. Food and office supply images are examples of photographs presented to children during fMRI.

Anticipated fullness was greater for larger amounts of food (EMM=51.4, SE=2.1) compared to smaller amounts of food (EMM=41.3, SE=2.1; ps<0.001; Figure 3C) and greater for higher-ED foods (EMM=49.6, SE=2.1) compared to lower-ED foods (EMM=43.1, SE=2.1; ps<0.001; Figure 3C). There was no interaction between ED and food amount on anticipated fullness (ps>0.23).

3.4. BOLD response to cue type

There was greater activation to food than office supplies in a cluster extending bilaterally through anterior cingulate cortex (ACC), medial prefrontal/orbitofrontal cortex (mOFC), frontopolar cortex and caudate, into left lateral OFC and left pallidum (Figure 4; Table 2). In addition, there was greater activation to food than office supplies in bilateral primary visual cortex (V1) and posterior insula extending into middle insula. There was lower activation to food than office supplies in bilateral clusters extending through inferior parietal cortex, visual cortices (V4, V7), ventromedial visual areas, fusiform face area, parahippocampal areas, and lateral occipital and posterior temporal cortices, as well as bilateral clusters extending from inferior frontal cortex to dorsolateral prefrontal cortex (dlPFC).

Figure 4.

BOLD response to cue type. Coordinates are in MNI space. Thresholded z-statistic map (voxel-wise p < 0.001, cluster-corrected to p < 0.05) of 1-sample t-test on cue type contrast (food - office supplies). Orange clusters show regions with greater responses to food cues than office supplies. Blue clusters show regions with greater responses to office supplies than food cues. Food and office supply images are examples of photographs presented to children during fMRI.

Table 2.

Regions showing BOLD response to cue type (food vs. office supplies)

Effect	Region	H	k a	Z b	x	y	z
Food > Office Supplies	Anterior cingulate cortex (ACC)	L	25,375	6.28	−5	41	3
	V1 / Inferior occipital gyrus	L	6,210	7.42	−17	−98	−10
	V1 / Lingual gyrus	R	5,328	7.84	19	−95	−9
	Posterior insula	L	3,664	7.97	−37	3	−10
	Posterior insula	R	2,877	7.31	37	5	−10
Food < Office Supplies	Inferior parietal cortex / middle occipital gyrus	L	52,337	−7.94	−27	−67	29
	Parahippocampal area	R	47,831	−7.76	30	−47	−8
	Inferior frontal sulcus / Prefrontal gyrus	L	16,484	−6.42	−47	0	32
	dlPFC / Inferior frontal gyrus	R	13,874	−5.93	45	25	22

Results from 1-sample t-test on cue type contrast (food - office supplies) controlling for average framewise displacement, sex, pre-MRI fullness, and pre-MRI anxiety (CAMS); Cluster p-values were cluster-corrected to p < 0.05

^acluster-extent in 1×1×1 mm³;

^bpeak cluster z-value derived from the general linear model

x, y, z: MNI coordinates of cluster peak; H: hemisphere; V1: primary visual cortex

3.5. BOLD response to ED

When larger and smaller food amounts were combined, there was greater activation to higher- than lower-ED foods in bilateral ACC/mOFC extending into bilateral dorsomedial prefrontal cortex and left ventromedial prefrontal cortex (vmPFC)/mOFC, frontopolar cortex, and lateral OFC (lOFC; Figure 5A; Table 3).

Figure 5.

BOLD responses to energy density (ED). Coordinates are in MNI space. Z-statistic maps were thresholded at voxel-wise p < 0.001 and cluster-corrected to p < 0.05. (A) BOLD response to ED: thresholded z-statistic map for 1-sample t-test on energy density contrast (higher - lower) across larger and smaller amount conditions. Orange clusters show regions with greater responses to higher-ED than lower-ED foods. No regions showed greater responses to lower-ED than higher-ED foods. Higher-ED and lower-ED food images are examples of photographs presented to children during fMRI. (B) ED combination map: z-statistic maps from separate 1-sample t-tests on ED contrasts for larger and smaller amount conditions were thresholded and binarized and then combined to show conjunction and disjunction in the extent of responses to ED. Colors reflect whether voxels exhibited responses to ED (higher vs. lower) for larger amounts (blue), smaller amounts (pink), or both smaller and larger amounts (purple).

Table 3.

Regions showing BOLD response to energy density (higher vs. lower)

Effect	Region	H	k a	Z b	x	y	z
BOLD response to ED across larger and smaller food amounts
Higher > Lower	ACC / medial OFC	L	13,492	5.49	−3	48	−7
BOLD response to ED for larger food amounts
High > Low	ACC / medial OFC	L	2,673	4.43	−3	48	−8
BOLD response to ED for smaller food amounts
High > Low	Lateral OFC	L	3,287	4.64	−24	59	−16
High < Low	Ventral visual complex / fusiform gyrus	L	2,091	−4.44	−36	−71	−16

Results from 1-sample t-tests on energy density (ED) contrast (higher - lower) controlling for average framewise displacement, sex, pre-MRI fullness, and pre-MRI anxiety (CAMS); Cluster p-values were cluster-corrected to p < 0.05

^acluster-extent in 1×1×1 mm³;

^bpeak cluster z-value derived from the general linear model

x, y, z: MNI coordinates of cluster peak; H: hemisphere

When analyzed individually by food amount, both smaller and larger food amount conditions showed greater activation for higher-ED relative to lower-ED foods in primarily distinct regions of bilateral ACC/mOFC, which overlapped in left ACC/mOFC (Figure 3B; Table 5). For smaller amounts only, activation extended laterally into left frontopolar and lOFC. Additionally, for smaller amounts only, there was less activation to images of higher- than lower-ED foods in fusiform face area, posterior inferotemporal complex, and lateral occipital cortex.

Table 5.

Regions showing differential BOLD responses to amount by cue type and energy density (ED)

Effect	Region	H	k a	Z b	x	y	z
BOLD response to amount (larger - smaller) by cue type (food vs. office)
Food > Office Supplies	V2 / Calcarine gyrus	L	4,838	7.24	−15	−94	−7
	V1 / Calcarine gyrus	R	4,078	6.83	18	−91	0
BOLD response to food amount (larger - smaller) by ED (higher vs. lower)
n.s

Results from paired t-tests using first-level amount contrasts (larger - smaller) controlling for average framewise displacement, sex, pre-MRI fullness, and pre-MRI anxiety (CAMS); Cluster p-values were cluster-corrected to p < 0.05

^acluster-extent in 1×1×1 mm³;

^bpeak cluster z-value derived from the general linear model

x, y, z: MNI coordinates of cluster peak; H: hemisphere; V1: primary visual cortex, V2: secondary visual cortex; n.s: no clusters surviving cluster-correction

3.6. BOLD response to amount

When higher-ED and lower-ED foods were combined, there was greater activation to larger than smaller food amounts in bilateral visual areas (V1–V4, V8, ventromedial visual area), and right parahippocampal area (Figure 6A; Table 4). There was less activation to larger than smaller food amounts in right angular gyrus.

Table 4.

Regions showing BOLD response to amount (larger vs. smaller)

Effect	Region	H	k a	Z b	x	y	z
BOLD response to food amount across higher- and lower-ED foods
Larger > Smaller	V3 / Lingual gyrus	L	53,596	13	−20	−88	−16
Larger < Smaller	Angular gyrus	R	2,416	−4.73	47	−56	31
BOLD response to food amount for higher-ED foods
Larger > Smaller	V2 / Inferior occipital gyrus	L	44,433	13	−16	−92	−9
BOLD response to amount for lower-ED foods
Larger > Smaller	V2 / Lingual gyrus	R	13,776	13	15	−91	−8
Larger > Smaller	V2/3 / Lingual gyrus	L	12,648	7.78	−16	−88	−11
BOLD response to amount for office supplies
Larger > Smaller	V3 / Lingual gyrus	L	73,549	13	−20	−88	−17

Results from 1-sample t-tests on amount contrasts (larger - smaller) controlling for average framewise displacement, sex, pre-MRI fullness, and pre-MRI anxiety (CAMS); Cluster p-values cluster-corrected to p < 0.05

^acluster-extent in 1×1×1 mm³;

^bpeak cluster z-value derived from the general linear model (maximum value output by AFNI is 13)

x, y, z: MNI coordinates of cluster peak; H: hemisphere; V1: primary visual cortex, V2: secondary visual cortex, V3: third visual cortex

When analyzed individually by image type, higher-ED food, lower-ED food, and office supply conditions all showed greater activation for larger relative to smaller amounts in V1–V4 and V8 (Figure 6B; Table 4). For higher-ED food and office supply conditions only, activation extended into additional bilateral ventromedial visual areas, right visual area V3B and right posterior parahippocampal area. For the office supply condition only, activation extended even further into right fusiform face area, bilateral posterior inferotemporal complex, bilateral parahippocampal areas, bilateral visual areas V7 and V3A, and bilateral inferior parietal cortex.

The BOLD response to amount (larger>smaller) was greater for food cues than office supplies in bilateral V1 and V2 (Figure 6C; Table 5) but did not differ by ED.

3.7. Associations between BOLD responses and percent wanting (%want)

BOLD responses to higher-ED foods, lower-ED foods, larger food amounts, smaller food amounts, and office supplies did not show significant covariation with %want parametric regressors, indicating that, within children, the intensity of the BOLD responses to each image condition was not significantly associated with the % of images that children reported wanting. However, the association between BOLD response and %want was weaker for higher-ED foods compared to lower-ED foods in right early visual cortex (V4; Table S3). The association between BOLD response and %want did not differ between food amount conditions.

3.8. Associations between BOLD responses, familial risk status and fat mass index

BOLD responses to cue type, ED, and amount were not significantly associated with familial risk status or fat mass index. Exploratory assessment of subthreshold results revealed higher fat mass index was associated with greater BOLD responses to higher- than lower-ED food cues in left dlPFC (p<0.06) and left V1 (p<0.09; Table S4).

3.9. Associations between BOLD responses and the PSE

In the cluster responsive to food amount that extended through visual and parahippocampal areas (peak coordinates = −20, −88, −16; Table 4), BOLD response to larger > smaller food amounts was negatively associated with the PSE for the weight of food consumed (stats; Table S11); the greater the BOLD response to larger > smaller food amounts, the smaller the increase in intake as portion sizes increased (Figure 7). This association was not observed for the PSE for energy consumption (Table S12). Food intake was not associated with BOLD responses in clusters responsive to cue type, ED, office supply amount, or the interaction between cue type and amount (Tables S13–S20).

Figure 7.

Interaction between BOLD responses to food amount (larger > smaller) and laboratory meal size condition on food intake. BOLD responses were extracted from a cluster extending through visual and parahippocampal areas that showed greater responses to larger > smaller food amounts in a 1-sample t-test (peak = −20, −88, −16). Regression lines for the association between meal size condition (i.e., proportion increase in size from reference portion; x-axis) and food intake in grams (y-axis) are plotted for three values of BOLD response to food amount: 1 SD below the mean (dotted line), the mean (dashed line), and 1 SD above the mean (solid line).

4. Discussion

Consistent with prior work showing independent effects of energy density and food amount (i.e., portion size; English et al., 2017; English et al., 2019), the present study found higher-ED food images elicited greater activation than lower-ED food images in regions associated with reward-related decision making and valuation while larger amounts of food elicited greater activation than smaller amounts in regions associated with visual processing. Extending prior work, we observed greater activation for higher- than lower-ED food cues in distinct regions of the OFC depending on whether foods were presented in larger or smaller (i.e., age-appropriate for children) amounts and identified different patterns of activation to amount in visual processing areas for foods and office supplies. These findings suggest brain responses to food ED vary depending on the amount of food presented, highlighting potential mechanistic avenues whereby the brain makes distinctions between food energy content and quantity.

Overall, there was greater activation for higher- than lower-ED foods across ACC, medial PFC, and OFC. However, BOLD responses to ED diverged for foods depending on whether they were presented in larger or smaller portion sizes. For both larger and smaller portions, BOLD responses to ED were found in the vmPFC (e.g., mOFC, ACC), whereas for smaller amounts only, the BOLD response to ED extended into lOFC. While both vmPFC and lOFC are implicated in food-related decision making (Rolls, 2021; Seabrook and Borgland, 2020; Zald, 2009) and food valuation (Motoki and Suzuki, 2020), functional differences have been reported between these regions. vmPFC is thought to encode overall value signals for food by integrating signals for multiple food attributes (e.g., taste, health) while lOFC encodes information about specific food attributes (Motoki and Suzuki, 2020), such as taste pleasantness (Small et al., 2003), sensory characteristics (Howard et al., 2015), and specific nutrients (Suzuki et al., 2017). In particular, a subregion of lOFC shown to encode perceptions of fat quantity in a food (Suzuki et al., 2017) overlaps with the subregion we showed was responsive to ED for smaller portions. While children may value higher-ED foods more than lower-ED foods regardless of portion, they may only incorporate perceptions of energy content or associated nutrients (e.g., fat) into valuations when portions are more age-appropriate (i.e., the smaller amount). Additionally, it is possible when less food is available for consumption, the brain prioritizes calculations of energy content to ensure biological adequacy of the meal. Future research should assess how perceptions about different food attributes (e.g., macronutrients, taste, health) are incorporated into value-based decisions for varying portions.

The BOLD response to ED also differed by food amount in regions implicated in object recognition and representation (Conway, 2018; Riesenhuber and Poggio, 2002); greater activation to lower- than higher-ED foods was observed in fusiform face and posterior inferotemporal areas for smaller but not larger portions of food. Fusiform has been shown to be more responsive to food and faces than other objects (Adamson and Troiani, 2018; Zheng et al., 2022), to food in a fasted than fed state (Holsen et al., 2005; LaBar et al., 2001; Uher et al., 2006), and to higher- than lower-energy foods (English et al., 2019; Masterson et al., 2015), suggesting motivational relevance (i.e., value) drives activation in this region (Adamson and Troiani, 2018). While we observed greater activation to lower- than higher-energy dense foods in this region, our finding aligns with previous work showing food attributes impact fusiform responses, particularly for smaller, age-appropriate portion sizes.

For food and office supply conditions, larger amounts were associated with greater activation in visual processing regions than smaller amounts. However, in primary and early visual cortices, the response to amount was stronger for food than office supplies, which may be due to differences in low-level visual characteristics (e.g., size, color, complexity) between these image conditions (Kling et al., 2020). Alternatively, this pattern could indicate children had greater value-based modulation of early visual cortex (Serences, 2008) for larger vs. smaller amounts of food compared to larger vs. smaller amounts of office supplies. Further, the BOLD response to amount was more widespread for office supplies than food items, extending bilaterally to regions associated with object recognition (e.g., fusiform; Riesenhuber and Poggio, 2002) and associative processing and episodic memory (e.g., parahippocampal regions; Aminoff et al., 2013). Given common school items were depicted, this response may reflect associative processes or memories related to experiences with larger amounts of office supplies in this context. Future research using other non-food categories can shed light on whether these distinct responses to amount for food vs. office supplies reflect differences in biological relevance or are due to previous experiences with the stimuli.

In contrast to our hypotheses, larger amounts of food did not elicit greater engagement in regions associated with reward or cognitive control compared to smaller amounts of food. This differs from a previous study in which we observed greater inferior frontal gyrus/OFC activation to larger vs. smaller portion sizes in children 7–10 years old (English et al., 2017). The younger age range and restricted weight status of the current sample may contribute to differences in findings. Further, as children’s susceptibility to the ‘portion size effect’ has been associated with food cue reactivity (English et al., 2019; Keller et al., 2018), weight status and appetitive traits (i.e., food and satiety responsiveness; Smethers et al., 2019) and executive functions (Keller et al., 2023), it is possible large portion size cues may only influence reward and cognitive processes, but only among some children (Keller et al., 2018). Alternativity, large portion size cues may modulate eating behaviors predominantly through perceptual processes rather than primary reward or cognitive mechanisms. In support of this, post-hoc analyses revealed that the average BOLD response to food amount (larger > smaller) in regions implicated in visual processing was negatively associated with the PSE measured across four laboratory meals. As the BOLD response to office supply amount (larger > smaller) in a similar, overlapping region was not associated with the PSE, results suggest a food-specific association where children who are more sensitive to visual portion size cues in brain regions associated with visual and associative processes are less susceptible to eating more as portion sizes increase. One possibility is that the inability to perceptually distinguish larger from smaller portions, whether implicit or explicit in nature, may make children more susceptible to the PSE because they are less aware that portions are increasing. Whether overtly telling children about the portion size manipulations would help to mitigate the effects of portion size on intake remains to be tested.

Also in contrast to our hypotheses, we found no differences in children’s BOLD responses to food cues by familial risk for obesity, defined using maternal BMI. This differs from prior studies in youth (Carnell et al., 2017; Luo et al., 2021) that observed associations between maternal BMI and food cue reactivity in regions associated with self-regulation, perceptual processes, and memory. Based on a comparison of methods and sample characteristics to studies by Carnell et al., (2017) and Luo et al., (2021), the lack of differences by familial risk status in the current study could be due to variations in physiological state, developmental stage, and/or demographic characteristics. Alternatively, loss of data due to motion may have left us underpowered to robustly test differences in food cue reactivity by familial risk and child adiposity; therefore, more work is needed to address these questions.

The present study has several strengths. First, we used a publicly-available, standardized set of images (Kling et al., 2020), which facilitates interpretation of results and replication in future studies. Second, children completed a two-session mock-MRI protocol prior to scanning, allowing them to become familiar with the MRI environment. Third, by using conjunction and interaction analyses, we were able to build on prior research examining BOLD responses to food ED and portion size (English et al., 2017) to provide insight into how foods at varying levels of energy density are distinctly processed when portions are smaller (age-appropriate) and larger. Fourth, by sharing raw data and unthresholded statistical maps, this study has the potential to contribute to future meta- and mega-analyses on food cue reactivity in children.

In addition to strengths, the present study has several limitations. First, our sample lacked racial and ethnic diversity. Future research is needed to assess whether results generalize to more diverse populations. Second, while office supplies were selected as the non-food condition because they are familiar to children and can be presented in varying colors/quantities, comparisons between food and office supplies may not generalize to comparisons between food and other non-food objects. For example, while prior studies have shown greater responses to food than non-food items (e.g., scenery, animals) in higher-order visual processing regions (see van der Laan et al., 2011 for meta-analysis), we observed the opposite pattern in comparisons between food and office supplies. Office supplies may be similar to tools in their ability to elicit greater activation in dorsal and ventral visual streams compared to other non-food categories (e.g., animals; Dekker et al., 2011). Future analyses comparing brain responses to amount between food and other non-food categories are needed in order to determine if there is a “food-specific” response to amount. Third, the binary wanting responses provided by children in the scanner may have limited our ability to detect whether BOLD responses are modulated by wanting. To better examine these associations, future studies could use a wanting scale that allows for more granularity in responses (e.g., a multi-point scale). Fourth, while we assessed in-scanner wanting and post-scan ratings of anticipated fullness and liking, it remains unclear how children perceived and interpreted the food cues during the scan. Future research could qualitatively investigate how children perceive images of large food portions or provide more explicit prompts to induce specific mental states (e.g., imagine you have to eat the full portion). Finally, some children categorized as having high familial risk for obesity in the present study may develop obesity and some may be resilient to obesity. Such heterogeneity could preclude the detection of brain phenotypes that promote adiposity gain in children. To better understand neurobiological traits that pre-dispose children to obesity, future analyses could stratify initial “at-risk” groups based on future weight gain and subsequently compare baseline food cue reactivity.

In conclusion, we showed that ED influenced BOLD responses in vmPFC for larger and smaller (i.e., age-appropriate) amounts of food, but influenced BOLD responses in lOFC and fusiform for smaller amounts only. Further, the response to amount was stronger and more localized in early visual processing areas for food images than non-food images, and that BOLD responses to food amount across visual and parahippocampal areas was associated with the effect of portion size on measured intake. These results suggest portion size impacts valuation processes related to food energy content and influences perceptual processes differently than non-food amount. These findings improve our understanding of how portion size impacts brain food cue reactivity in children and elucidate potential neurobiological mechanisms underlying the “portion size effect.”

Data availability

Data and analytic code described in the manuscript are available on Open Science Framework (https://osf.io/x8j56/) and OpenNeuro (https://openneuro.org/datasets/ds004697/versions/1.0.2)