Introduction

Food fussiness (also known as picky eating) in children is characterized by the rejection of a large proportion of familiar and unfamiliar foods, leading to decreased consumption of food variety1. The prevalence of food fussiness is believed to be around 22%, with food fussiness increasing with age in young children2. Picky eating has a well-documented association with decreased consumption of vegetables in preschoolers and children3,4,5,6,7,8. Picky eating has also been associated with lower enjoyment of fruit6 and lower fiber intake7and demonstrates an indirect association with greater ultra-processed food intake in children9. Picky/fussy eating may lead to negative health consequences due to the lack of essential nutrients resulting from low vegetable, fiber, and fruit intake10. Further, in the KOALA birth cohort, picky eaters were more likely to be shorter and more often underweight in comparison to non-picky eating peers at five years old11.

Childhood food fussiness often resolves with age10. However, some individuals maintain picky eating into adolescence12and picky eating in childhood may increase the risk of developing eating disordered behaviors in adolescence10,13. Further, while common in childhood at a moderate level, picky eating may sometimes form part of a broader clinical feeding disorder, e.g. pediatric feeding disorder or Avoidant Restrictive Food Intake Disorder [ARFID] that requires therapeutic and sometimes medical intervention14. While picky eating is often associated with healthy or low weight, decreasing food fussiness in preschoolers with obesity has been found to aid in healthier dietary changes15. Understanding and overcoming fussy eating may therefore help to promote a healthy diet and eating habits in individuals across the weight spectrum1, as well as to potentially decrease the risk of disordered eating.

Hypothesized contributors to picky eating include both genetic factors16,17,18 and environmental factors such as parental feeding practices10,19,20 and exposure to novel foods16. Further, a small body of neuroimaging research is beginning to illuminate neural underpinnings of picky eating. For example, a study of 10y olds found that those demonstrating symptoms of ARFID including picky eating exhibited greater cortical thickness in frontal and superior frontal regions21, while another found that food fussiness was associated with greater cortical thickness in the rostral middle frontal gyrus from 6 to 8 years, but with lesser cortical thickness in the same region from 12 to 14 years of age22. Typical brain development involves a regionally-variable process of initial growth and proliferation of synapses that peaks around age 7y and is reflected by an increase in cortical thickness, followed by a process of synaptic pruning that increases efficiency and manifests in cortical thinning23. Previously observed patterns of frontal cortex growth could therefore reflect altered development within regions subserving goal-directed behavior, some of which have been implicated in neurodevelopmental differences24. However, these studies did not investigate differences in brain functioning, which may be more directly relevant to the behavioral phenotype of picky eating. Further investigation of functional as well as structural neural correlates of food fussiness could expand understanding of its neural underpinnings and potentially help identify neural signatures that could be used as objective markers of treatment success. Individualized or general education about brain contributions to food avoidant behavior could also be explored as a means to increase understanding and diffuse anxiety around fussy eating, potentially aiding families in clinical or non-clinical settings.

Functional Magnetic Resonance Imaging [fMRI] food cue reactivity tasks have been used to understand neural correlates of individual differences in appetitive characteristics with known associations with body weight in a variety of populations including adolescents and younger children25,26,27,28 as well as to investigate the neural correlates of obesity or markers of obesity risk29,30. Such tasks often compare neural responses to photos of foods selected to be relatively high in energy-density with photos of foods selected to be comparatively low in energy-density, often including fruits and vegetables27,28,31. However, we are not aware of any published studies that attempt to characterize neural responses to foods known to be disliked in individuals with picky eating (e.g. fruits and vegetables), among individuals known to be high in picky eating.

For the current study we therefore leveraged a dataset including task fMRI data on neural responses to fruit and vegetable stimuli and measures of laboratory-assessed and free-living food intake25 to conduct a preliminary investigation of neural, as well as behavioral, correlates of food fussiness. Participants completed the fMRI visual food cue paradigm in a fasted state, followed by a multi-item ad libitum meal [ALM] and eating in the absence of hunger [EAH] task, and parents completed the Child Eating Behavior Questionnaire [CEBQ]32, a widely used parent-report questionnaire to measure appetitive traits in youth that contains a food fussiness subscale often used to assess picky eating behaviors, and which has a validated cut-off score for clinically significant food fussiness33. We hypothesized that individuals who were high vs. low in food fussiness would report decreased relative intake of fruits and vegetables, in parallel with increased relative intake of palatable high energy-density foods, on laboratory and free-living measures. We further hypothesized that the high vs. low food fussiness group would show differential patterns of brain activation in response to fruit and vegetable stimuli, as well as to stimuli representing palatable high energy-density foods.

Methods

Participants

Adolescents (14–18 years old) and their biological mothers were recruited as previously described as part of a cross-sectional study of familial obesity risk27,34,35 with written informed consent and assent obtained from mother and child. Exclusion criteria for both mother and child were assessed using brief screening questionnaires asking for self-report information including on health status, treatments including medications, and behaviors. Exclusion criteria for adolescents included reports of previous or current diagnosis of a significant health problem (including diabetes), psychiatric condition (including eating disorders, e.g. anorexia nervosa, bulimia nervosa, binge eating disorder), developmental disorder (e.g. autism, ADHD), or neurological disorders (e.g. epilepsy, head injury, history of concussion), use of medication that affects body weight or appetite (e.g. stimulants, anti-depressants), medical contraindications to an MRI, food allergies, or participation in a structured weight loss program. Exclusion criteria for mothers included heavy smoking, current pregnancy, recreational drug use, or high alcohol intake. This study was approved by the Johns Hopkins University School of Medicine Institutional Review Board and all methods were performed in accordance with the relevant guidelines and regulations.

Procedure

Study procedures are summarized in Fig. 1. Anthropometric measures were collected by research staff during an initial consultation visit. Height was measured using a wall-mounted stadiometer, and body weight and fat percentage were assessed using a SC-331S Total Body Composition Analyzer (TANITA Corp., Tokyo), which measures body weight and estimates fat percentage via Bio-Impedance Analysis. Mothers completed the Child Eating Behavior Questionnaire [CEBQ]32 and a questionnaire assessing sociodemographic variables including race, education, and household income. Adolescents also completed the Automated Self-Administered 24 h [ASA24] dietary assessment. A further 3.5 h afternoon visit (scan day) took place in a fasted condition. Adolescents were instructed to fast for 4 h prior to this visit. On arrival at the lab, the adolescents consumed a 474 ml glass of water. Thirty minutes afterwards, adolescents underwent an fMRI visual food cue reactivity task, followed by a multi-item ad libitum meal [ALM], and then an eating in the absence of hunger [EAH] task. On a separate day, adolescents underwent a similar set of procedures following a milkshake preload (474 ml, 480 kcal, fed condition), with the order of conditions counter-balanced across participants. Data from the fed condition were not included in the current analysis but are described and correlated with other CEBQ subscales elsewhere27,34,35.

Fig. 1
figure 1

Study schematic. Notes: fMRI = functional Magnetic Resonance Imaging; ED = Energy-Density; F&V = fruit and vegetable; ASA24 = Automated Self-Administered 24 h [ASA24] dietary assessment.

Full size image

Measures

Child eating behavior questionnaire [CEBQ]

The CEBQ32 is a 35-item parent-report questionnaire assessing multiple dimensions of food approach and food avoidance with demonstrated correspondence with behavioral tests of appetite36 and robust associations with body weight37. The measure contains a total of 8 subscales, with 4 assessing food approach traits (i.e., Food Responsiveness, Enjoyment of Food, Emotional Overeating and Desire to Drink) and 4 assessing food avoidance traits (Satiety Responsiveness, Slowness in Eating, Emotional Undereating and Food Fussiness). The response scale for each item included the options: Never, Rarely, Sometimes, Often, and Always. Scores for each subscale were calculated as item means ranging from 1 to 5. For this study our primary measure was the Food Fussiness sub-scale, containing the following items: ‘My child refuses new foods at first’, ‘My child enjoys tasting new foods’ (reverse scored), ‘My child enjoys a wide variety of foods’ (reverse scored), ‘My child is difficult to please with meals’, ‘My child is interested in tasting food s/he hasn’t tasted before’ (reverse scored), ‘My child decides that s/he doesn’t like a food, even without tasting it’. The Food Fussiness sub-scale has been validated against behaviorally coded videos of mother-child meals38. The internal consistency for the food fussiness subscale in our sample as assessed by Cronbach’s alpha was 0.86.

Laboratory meal tests

Ad libitum meal [ALM]

A multi-item ad libitum buffet meal was administered after the scan and set up as depicted in27. This meal included three Freschetta Natural Rising 12″ pizzas cut into 12 slices: one plain cheese pizza (790 g, 2009 kcal), one vegetable (peppers, tomatoes and onions) pizza (940 g, 2050 kcal), and one pepperoni pizza (825 g, 2199 kcal), Sabra hummus (283 g, 700 kcal), Kraft Classic Ranch dressing (224 g, 880 kcal), Breyer’s Homemade Vanilla ice cream (250 g, 530 kcal), Chips Ahoy chocolate chip cookies (200 g, 970 kcal), Entemann’s Little Bites Fudge Brownies (350 g, 1200 kcal), M&Ms (200 g, 1000 kcal), Ruffles potato chips (200 g, 1143 kcal), Cheetos (250 g, 1339 kcal), baby carrots (250 g, 103 kcal), cherry tomatoes (300 g, 55 kcal), celery sticks (200 g, 32 kcal), grapes (600 g, 414 kcal), water (20 fl. oz, 0 kcal), regular Coke (20 fl. oz, 240 kcal) and diet Coke (20 fl. oz, 11 kcal). Adolescents were informed they would be left alone for 30 min and to eat as much as they wanted. Participants were told to consider the meal their dinner, and to therefore imagine that they would not eat anything for 4–5 h following the meal. Participants could step out of the room if they finished eating early or request extra time. Food was weighed prior to and after the meal to determine the amount consumed. Food was grouped into 4 categories for analysis: pizza, snacks, condiments, and fruits & vegetables.

Eating in the absence of hunger [EAH] test

After the ALM, adolescents were brought into a separate room in which an iPad and snacks were available on the desk. The snacks consisted of the following items: Doritos chips (250 g, 1248 kcal), Pringles (150 g, 803.6 kcal), Mini Oreos (150 g, 714.3 kcal) and Donut holes (200 g, 859 kcal). Adolescents were informed that the research assistant needed to prepare for the last task of the day, so they were free to use the iPad (containing games and internet access with family safety settings activated) and to eat as many of the snacks as they wanted. Adolescents were left in the room for 15 min. Snacks were weighed pre- and post- test to determine the amount consumed.

Participants were asked how hungry and how full they felt on a scale from 0 (not at all) to 100 (extremely) immediately before and after the ALM (i.e. just before the EAH test).

Free-living dietary intake

Automated self-administered 24-hour dietary recall [ASA24]

Participants completed one weekday Automated Self-Administered 24 h [ASA24] dietary assessment39,40 during the initial consultation visit. The ASA24 is an automated multiple-pass recall where participants record their diet and food intake from the previous day34. With the help of the research assistant, participants input the foods they ate, portion sizes and food preparation information. The ASA-2014 and ASA-2016 versions were used in this study. The ASA24 program utilizes the USDA’s Food and Nutrient Database for Dietary Surveys to generate nutrient intake and food group information for each participant. We examined generated variables representing added sugar, total fat (i.e. saturated, monounsaturated, polyunsaturated), protein, total carbohydrates, and empty calories (i.e. sum of added sugar, solid fat, and alcohol beyond energy-adjusted moderate amounts), as a percentage of overall intake.

Neuroimaging protocol

Food cue reactivity task

The food cue reactivity task, programmed in E-Prime (Psychology Software Tools, Inc., Sharpsburg, Pennsylvania), included 2 runs with 45 trials each (15 high energy-density [High-ED] food trials, 15 fruits and vegetables [F&V] trials, and 15 Non-food trials). Trials were pseudorandomized with no more than 2 stimuli of each category being presented in a row. Each trial lasted 4 s (including stimulus + stimulus rating period) followed by a fixation period (central crosshair for 1 s). For food stimuli the rating question was: Do you want to eat this right now? For Non-food stimuli the rating question was: Do you want to use this right now? Response choices at the bottom of the screen were: Not at all, Not really, A little bit, Very much. Participants selected responses with button boxes using two fingers (index, middle finger) on each hand. During the initial consultation visit, adolescents completed familiarity (Do you know what this is? with response options: No, Yes), frequency of consumption (Do you ever eat/use this food/object? with response options: 1 = Never, 2 = Rarely, 3 = Sometimes, 4 = Often) and liking ratings (How much do you like this food/object? with response options: 1 = I dislike it extremely, 2 = I dislike it, 3 = Neutral, 4 = I like it, 5 = I like it extremely) for all food and Non-food stimuli. For frequency of consumption and liking ratings, mean scores for the High-ED, F&V, and Non-food stimuli type were created. For Familarity, percentages for each stimuli type were created by recording how many items adolescents knew within each category (e.g. knowing 14 out of the 15 F&V items). Food and Non-food stimuli were colored photos approximately matched on size, shape, color and resolution. High-ED food photos included examples such as slice of pizza, chocolate chip cookies, ice cream, and M&Ms. F&V photos included examples such as green grapes, strawberries, and baby carrots. Non-food photos included examples such as a highlighter, masking tape, and gold screws. Participants were asked how hungry and how full they felt on a scale from 0 (not at all) to 100 (extremely) immediately before and after the scan.

Image acquisition

Images were acquired on a Philips 3T Achieva Multix X-Series scanner with 32-channel SENSE head coil. Functional images were obtained using a T2*-weighted single-shot echo planar imaging [EPI] pulse sequence with parameters as follows: repetition time [TR] = 2500 ms, echo time [TE] = 30 ms, in-plane resolution of 3.05 × 3.15 mm, field of view [FOV] of 256 mm × 256 mm, slice thickness 3 mm with no slice gap, and a flip angle of 70°. Functional volumes consisted of 47 slices, acquired in an ascending order and oriented along the AC-PC line.

Statistical analysis

Following33, we determined food fussiness groups based on CEBQ Food Fussiness subscale scores with a high food fussiness (high FF) group defined as having an item mean score greater than or equal to 3.00, which corresponds to high or moderate levels of picky eating based on clinical interview. In order to maximize phenotypic differences and clinical interpretability for this preliminary investigation, as well as to achieve comparable group sizes for fMRI analyses, we further defined a low food fussiness (low FF) group with CEBQ Food Fussiness scores less than or equal to 2.00, and did not further consider the intermediate, moderate group scoring between 2 and 3. BMI (kg/m2), BMI z scores and percentiles based on CDC reference data41 were calculated. Independent samples t-tests and chi-square tests were used to compare FF groups on continuous and categorical variables, respectively. Objective laboratory meal test data (total intake, pizza intake, snack intake, condiment intake and fruit and vegetable intake for ALM; total intake for EAH) were reported as absolute kcal values and as percentages of total kcal intake. From the dietary recall we analyzed variables representing added sugar, total fat, protein, total carbohydrates, and empty calories as a percentage of overall intake since absolute values from such recall data are less reliable42. Wanting ratings within the food cue reactivity task were used to create mean scores for each stimulus type, as well as High-ED–Non-food and F&V–Non-Food difference scores. Similarly, mean scores were created for Liking and Frequency of Consumption ratings, and Familiarity responses were summarized as percentages.

Neuroimaging processing was conducted as previously described34 using SPM12 (http://www.fil.ion.ucl.ac.uk/spm/), executed in MATLAB 2017a. The Artifact Detection Tools (ART) toolbox for MATLAB (Gabrieli Laboratory, MIT, Cambridge, MA) was utilized to identify global mean and motion outliers. Participants with more than 20% of volumes tagged as high motion (motion > 3 mm in any direction) were excluded from the analysis. For the remaining participants, functional images were first corrected for slice timing differences using the median slice as the reference. The images were then realigned to the mean image, normalized to the EPI template within SPM12, and spatially smoothed using a Gaussian smoothing kernel with a full width at half maximum (FWHM) of 6 mm. In the first-level analyses, task-related activity for each participant was detected by applying the general linear model (GLM) to the data, using the canonical hemodynamic response function (HRF) convolved with a boxcar function (BCF) representing the onsets and durations of High-ED food cues, F&V food cues, and Non-food cues. The six realignment motion parameters generated during preprocessing were included as covariates. The model estimation was performed using the Restricted Maximum Likelihood (ReML) algorithm, and task-related T contrast images (High-ED vs. Non-food, F&V vs. Non-food) were generated. For the second-level models, we assessed the random effects of task-related activity between groups (high FF vs. low FF groups) using Bayesian posterior inference applied to the contrast images from the first-level analysis. The Bayesian approach estimates the posterior probability of the observed group effects given the activation map without requiring strong assumptions about effect size43 thus avoiding the risk of over-correction via multiple comparison adjustment44,45. Clusters were reported if they had an effect size Cohen’s d > 0.2, a Bayes factor logBF > 3.0, and a cluster extent threshold k ≥ 4.

We present unadjusted analyses rather than analyses adjusting for factors such as age, sex and BMIz for this preliminary investigation, because these factors did not significantly differ between FF groups. Further, we were primarily concerned here with proximal neural correlates of FF regardless of potential upstream predictors of FF and its neural and behavioral consequences, such as age and sex, and potential downstream consequences, such as BMIz. We additionally note while child age and sex could theoretically influence FF, differences in outcomes by age and sex do not confound the direct relationship of interest which is between FF and neural and behavioral outcomes. For example, if younger age and male sex were associated with increased FF, we propose that this variation should be captured by variation in FF scores and thus those factors should contribute to rather than obfuscate variance in the neural and behavioral outcomes of interest.

Results

Sample characteristics

Our total analysis sample included adolescents with complete data on demographic characteristics and CEBQ subscales who met criteria for low and high FF, amounting to n = 28 in the low FF group and n = 23 in the high FF group. Following exclusions for unavailability of imaging data or unacceptable motion parameters, our neuroimaging sub-sample comprised n = 19 in the low FF group and n = 17 in the high FF group.

Characteristics for the total analysis sample are presented in Table 1. Participants had a mean age of 16.23 ± 1.15 years ranging from 14.39 to 18.94. Twenty-nine participants were male and 22 female. Mean BMI z-score was 0.26 ± 1.20 with 35 participants classified as having healthy-weight, 9 as having overweight, and 7 as having obesity. Based on mother-reported race, 27 participants were white, 16 were black or African American, 2 were Asian, 5 were more than 1 race, and 1 unreported. Mean maternal BMI was 28.18 ± 7.97 with a range from 18.8 to 60.1. As previously described, a socioeconomic disadvantage index was calculated based on income, education, food insecurity and receipt of public assistance46 ranging from 0 to 7 with higher scores indicating more disadvantage. The overall sample scored 2.16 ± 2.02. Adolescent age, BMI-z, maternal BMI, adolescent weight groups, sex, race and socioeconomic disadvantage, did not differ between groups. On the CEBQ, as expected, the high vs. low FF group had lower Enjoyment of Food and higher Satiety Responsiveness.

Table 1 Sample characteristics for low food fussiness (Low FF) and high food fussiness (High FF) groups.
Full size table

Further, subjective ratings of hunger and fullness ascertained immediately before and after the scan, and immediately before and after the ALM (i.e. just before the EAH test), did not differ between FF groups (Supplementary Table 1).

Laboratory meal tests: ALM and EAH

The low and high FF groups did not differ on total ALM or EAH intake in kcal, or on absolute intake of pizza, snacks, or condiments. However, the high vs. low FF group had lower absolute levels of fruit and vegetable intake (41.73 ± 135.88 kcal vs. 86.12 ± 69.50 kcal, p = .010) (see Fig. 2 and Supplementary Table 2).

Fig. 2
figure 2

Ad libitum Meal fruit and vegetable intake in calories by Food Fussiness group. Notes: Food Fussiness groups were compared on fruit and vegetable intake in calories using independent samples t-tests. FF = Food Fussiness; * p < .05.

Full size image

ALM intake by food group expressed as a percentage of total intake is depicted in Fig. 3. The high vs. low FF group had a lower percentage fruit and vegetable intake (3.18 ± 2.59% vs. 7.22 ± 7.94%, p = .027) (Fig. 3). There were no significant group differences in intake of pizza, snacks, or condiments expressed as a percentage of total intake (Supplementary Table 2).

Fig. 3
figure 3

Ad Libitum Meal intake by food group as percentage of total caloric intake by Food Fussiness group. Notes: Food Fussiness groups were compared on fruit and vegetable intake in calories as a percentage of total meal intake in calories using independent samples t-tests. FF = Food Fussiness; * p < .05.

Full size image

Mean duration for the ALM task was 23 min and did not differ by FF group (high FF 23 ± 7 min, low FF 23 ± 6 min).

Free-living dietary intake: ASA24

Analyses of ASA24 data revealed that the high FF vs. low FF group consumed a greater percentage of total caloric intake from added sugar (13.53 ± 7.86% vs. 9.06 ± 5.69%; p = .032). There were no significant group differences in intake of total fat, protein, total carbohydrates, and empty calories expressed as a percentage of total intake. See Supplementary Table 2 for full details.

fMRI protocol

Wanting scores

Wanting ratings within the food cue paradigm are depicted in Fig. 4 with full details presented in Supplementary Table 3. The high vs. low FF group had lower wanting scores for F&V (2.44 ± 0.50 vs. 2.88 ± 0.51, p = .004). The high vs. low FF group additionally had lower wanting scores for Non-foods (1.47 ± 0.45 vs. 1.80 ± 0.51, p = .020). Groups did not differ in absolute wanting scores for High-ED stimuli. However, in analyses of difference scores that mapped more directly onto the contrast analyses presented for the neuroimaging data, the high vs. low FF group had higher High-ED-Non-food difference scores (1.47 ± 0.65 vs. 1.07 ± 0.56; p = .028), reflecting a greater relative preference for High-ED foods to Non-foods.

Fig. 4
figure 4

Wanting scores for High-ED, F&V and Non-food stimuli in food cue reactivity task by Food Fussiness Groups. Notes: Food Fussiness groups were compared on mean wanting ratings for (A) High energy-density foods, (B) Fruits and vegetables, and (C) Non-food stimuli using independent samples t-tests. ED = Energy-Density; FF = Food Fussiness; * <0.05, **<0.01.

Full size image

Liking, familiarity and frequency of consumption

The high vs. low FF group additionally had lower liking ratings (3.32 + 0.42 vs. 3.83 + 0.38; p < .001), lower familiarity (95.2 + 5.9% vs. 98.6 + 2.9%; p = .011) and lower frequency of consumption (2.39 + 0.38 vs. 2.77 + 0.31; p < .001) for F&V stimuli. There were no differences in liking, familiarity or frequency of consumption between FF groups for High-ED foods or Non-foods (See Supplementary Table 4).

Neuroimaging results

Comparison of high FF and low FF groups

High-ED vs. Non-food contrast

The high FF group demonstrated greater activation in cerebellum, posterior orbital gyrus, and a small cluster within the putamen (k = 4) compared to the low FF group, driven by a combination of greater activation in response to High-ED foods vs. Non-foods within the high FF group and lesser activation (i.e. relatively greater activation to Non-foods vs. High-ED foods) within the low FF group. (See Fig. 5a for activation maps showing group differences, Table 2 for full details on cluster size, location and Supplementary Fig. 1a for bar graphs of corresponding parameter estimates).

Table 2 High-ED foods vs. Non-foods.
Full size table
Table 3 Fruits & vegetables vs. Non-foods.
Full size table
Fig. 5
figure 5

(a) Areas demonstrating differential activation between Food Fussiness groups during food cue reactivity fMRI task in response to High-ED foods vs. Non-foods. Notes: Brain activation maps depict results from High FF vs. Low FF group comparison for the high-ED food vs. non-food contrast (Cohen’s d > 0.2, a Bayes factor logBF > 3.0, k ≥ 4). Red indicates greater activation in High FF vs. Low FF group in response to high-ED foods vs. non-foods; blue indicates lower activation in High FF vs. Low FF group in response to high-ED foods vs. non-foods. ED = Energy-Density; FF = Food Fussiness. (b) Areas demonstrating differential activation between Food Fussiness groups during food cue reactivity fMRI task in response to Fruits & Vegetables vs. Non-foods. Notes: Brain activation maps depict results from High FF vs. Low FF group comparison for fruits and vegetables vs. non-food contrast (Cohen’s d > 0.2, a Bayes factor logBF > 3.0, k ≥ 4). Red indicates greater activation in High FF vs. Low FF group in response to fruits and vegetables vs. non-foods; blue indicates lower activation in High FF vs. Low FF group in response to high-ED foods vs. non-foods. ED = Energy-Density; FF = Food Fussiness.

Full size image
F&V vs. Non-food contrast

The high FF group demonstrated lesser activation of the superior parietal lobule, central operculum, superior parietal lobule, precentral gyrus, and superior temporal gyrus, driven by a combination of greater activation in response to F&V vs. Non-foods within the low FF group and lesser activation (i.e. relatively greater activation to Non-foods vs. F&V) within the high FF group. The high FF group additionally demonstrated greater activation in a small cluster within the angular gyrus (k = 4) compared to the low FF group, driven by lesser activation among the low FF group. (See Fig. 5b, Table 3and Supplementary Fig. 1b).

Discussion

Our novel investigation of neural and behavioral correlates of a parent-report measure of food fussiness in adolescents demonstrated a pattern of differential brain response to food cues in adolescents with high vs. low FF that was consistent with complementary objective and reported intake data collectively demonstrating decreased approach toward F&V and increased approach toward palatable High-ED foods. These results suggest that food fussiness may in part be driven or sustained by altered patterns of engagement within a distributed circuit of brain regions implicated in sensorimotor processing and taste reward as well as higher-order cognitive processing, on visual exposure to foods.

Our behavioral results were largely consistent with previous research on food fussiness, also termed picky eating. For example, the high vs. low FF group ate less fruit and vegetables in absolute terms and relative to total intake as objectively measured at a laboratory-administered multi-item ad libitum buffet meal. We further note that this seems to match their usual consumption patterns for the specific F&V offered in our meal, as they reported lower free-living frequency of consumption (p = .018, data not reported). This extends similar findings of previous literature predominantly using self-reported data from 24-hour dietary recalls, food frequency questionnaires, and food diaries3,4,6,7,8,47.

Further, consistent with other literature suggesting poorer dietary quality among individuals high in food fussiness10,47, analyses of 24-hour dietary recall questionnaire data revealed that the high vs. low FF group consumed a higher percentage of their daily calories from added sugar, likely reflecting higher consumption of high-sugar snack-foods. It was notable that this pattern was not observed for percentage of daily calories from fat or protein, likely because preferences for animal foods tend to be lower in those with picky eating47. Although BMI z-scores were numerically lower in the high vs. low FF group, groups did not show a significant difference. While this finding is somewhat inconsistent with previous findings of lower weight in picky eaters11, it is consistent with our behavioral data showing that total caloric intake in both the ALM and EAH were equivalent between groups. We also note that while FF did not result in lower BMI in the current sample, those with high FF may still have been at risk of ingesting lower amounts of essential vitamins and nutrients found in fresh produce. Importantly we do not believe that the behavioral differences we observed were driven by age or sex as we saw no differences between groups.

Application of Bayesian methods to neuroimaging data acquired during our food cue reactivity task yielded several notable observations. First, visual exposure to palatable high-ED foods (vs. Non-foods) elicited greater activation within the high vs. low FF group in the posterior orbital gyrus and cerebellum, as well as in the putamen (although this was within a very small cluster of only four voxels). These regions are implicated in food including taste reward as well as motor function and emotional processing29. Consistent with these results, wanting ratings for High-ED vs. Non-food stimuli made during their presentation during the scan were higher in the high FF group. In contrast, visual exposure to F&V (vs. Non-foods) elicited an extensive pattern of lesser activation in the high vs. low FF group across multiple brain areas including regions involved in taste (central operculum), as well as aspects of object processing (superior temporal gyrus, superior parietal lobule). We also observed lesser activation in the right precentral gyrus, a premotor region, although this could in part be an artefact of positioning of the button boxes for ‘Not at all’ and ‘Not much’ responses to the wanting rating in the left hand, and should therefore not be overinterpreted. Importantly, this global pattern of lesser activation was accompanied by evidence for lesser absolute and relative F&V intake at a consequent ad libitum meal in the high FF group. Further, wanting ratings for F&V stimuli made during their presentation during the scan were lower in the high FF group. We therefore speculate that the group differences in neural activation that we observed may indicate an overall pattern of relatively lesser engagement with and interest in F&V among adolescents with high FF.

We acknowledge several limitations to our study, one of the most important of which is the small sample size and restricted variation in food fussiness in this community sample, which may have limited our ability to identify small and/or reliable effects. Post-hoc power calculations for this secondary analysis suggested that our power to detect a difference between high and low FF for ad libitum F&V intake was low, which could mean that this result was attributable to Type I error. However given that the finding is consistent with previous literature and was obtained using objective intake data from a standardized multi-item meal administered in laboratory conditions with control for nutritional status (fasted state), we are more inclined to think that the behavioral effect we detected is real, and large. Due to the lack of previous neuroimaging data on this topic we adopted a Bayesian whole-brain approach and post-hoc power analyses are not readily calculated. Our neuroimaging results in this small sample should therefore be considered preliminary and require replication in an independent dataset. Another significant limitation is that although we excluded individuals who reported previous or current psychiatric diagnoses (e.g. eating disorders such as AN, BN and BED) or developmental disorders (e.g. autism, ADHD), we did not conduct a comprehensive psychiatric evaluation to allow detection of these potential confounders, or for other conditions or traits that are associated with food fussiness and may therefore be considered as confounders, e.g. anxiety, obsessive compulsive disorder [OCD], sensory sensitivities8,48,49. A larger dataset with better characterization of these potential confounders would enable a more rigorous investigation of variation in food fussiness throughout the distribution. Relevant to the issue of potential confounding, although Non-foods were included as a neutral control stimulus, we saw some evidence for lesser wanting of Non-foods in the high FF group suggesting the presence of behavioral differences not specific to food. However, our neuroimaging contrast analyses and wanting difference score analyses accounted for this by subtracting out Non-food effects. A further caveat is that because we saw differences between our FF groups on other appetitive dimensions, namely enjoyment of food and satiety responsiveness, we also cannot conclude that the phenomena we observed were specific to food fussiness independent of other food avoidant behaviors.

A strength of our study was our multimodal behavioral protocol which lends validity to the neuroimaging results by effectively phenotyping food fussiness by a combination of parent-report questionnaire, child-report free-living dietary intake, and objectively measured laboratory intake as well as stimulus-specific child-reported wanting ratings. However we also note that eating in a laboratory setting may have altered (possibly decreased) intake in some individuals50 such that our results may not be a true reflection of free-living dinner meal intake. The 24 h dietary recall is also limited and does not provide information about longer-term dietary patterns including patterns earlier in childhood, which would provide greater evidence that we were indeed capturing a severe or enduring fussy eating profile. We additionally note that the results we observed were obtained using a food cue task that incorporated a ‘wanting’ rating and our findings may therefore be specific to neural correlates of evaluation of foods rather than passive responses. Further, our task design grouped food cues into High-ED and F&V categories but there is certainly substantial individual variation within these categories such that not all foods within the category are equally liked or disliked by all participants. Averaging activation within category and within group is therefore likely to underestimate the impact of food fussiness on neural responses in comparison to a more individualized approach.

In conclusion, we add to a small body of work demonstrating structural brain differences in association with picky eating in children21,22 to show that food fussiness in adolescents is associated with differential patterns of brain activation elicited by visual cues representing fruits and vegetables and highly palatable high energy-density foods relative to non-foods, which are accompanied by a behavioral phenotype of lower consumption of F&V ascertained by both objective and self-report methods. Future research directions to investigate the biological including neural underpinnings of food fussiness could include evaluating the contribution of sensory sensitivities as seen in neurodevelopmental conditions including autism and avoidant/restrictive food intake disorder (ARFID), as well as of other relevant psychiatric comorbidities, and genetically-driven biological variations such as supertaster status51. Longitudinal data collected from early life is needed to illuminate neural predictors and potential mechanisms driving development of food fussiness, and interventional research could help determine whether factors known to be protective against food fussiness (e.g. involving children in food preparation48 or the family eating the same meal as the child10) could impact the neural circuits underlying food fussiness. Further work combining both structural and functional assessments with behavioral outcomes is also needed to establish the consequences of the brain differences reported here and in other studies. However, our extant findings suggest that the response patterns might have value as objective markers of treatment success. Individualized or general education about the role of these brain circuits in food avoidant behavior might also be a useful component of clinical assessment or community-level information to help families struggling with food fussiness.