Trait food craving predicts functional connectivity between dopaminergic midbrain and the fusiform food area during eating imagery

Neurofunctional coupling between the dopaminergic midbrain (i.e., ventral tegmental area, VTA) and higher-order visual regions may contribute to food craving, leading to the onset or maintenance of obesity. We recently showed that the VTA resting-state functional connectivity with the occipitotemporal cortex, at the level of the fusiform gyrus (FFG), was specifically associated with trait food craving and the implicit bias for food images, suggesting that VTA-FFG connectivity may reflect the association between the visual representations of food and its motivational properties. To further test this hypothesis, this time we studied task-based functional connectivity in twenty-eight healthy-weight participants while imagining eating their most liked high-calorie (HC) or least liked low-calorie food (LC) or drinking water (control condition). Trait food craving scores were used to predict changes in task-based functional connectivity of the VTA during imagery of HC compared to LC foods (relative to the control condition). Trait food craving was positively associated with the functional connectivity of the VTA with the left FFG: people with higher trait food craving scores show stronger VTA-FFG connectivity, specifically for the imagery of the liked HC foods. This association was not linked to the quality of imagery nor to state measures of craving, appetite, or thirst. These findings emphasize the contribution of the functional coupling between dopaminergic midbrain and higher-order visual regions to food craving, suggesting a neurofunctional mechanism by which the mental representations of the HC food we like can become much more salient if not irresistible.


Supplementary methods 1.Participants
Participants were required to fast for 4-5 hours prior to the experiment and to avoid drinking beverages other than water.Upon their arrival, all participants completed a screening for eating disorders, including anorexia nervosa, bulimia nervosa, and binge-eating disorder.In particular, participants were presented with: (i) the Binge Eating Scale (BES) 1 , a measure constructed to described behavioral manifestations and feelings/cognitions surrounding a binge episode; and (ii) the Eating Disorder Inventory (EDI-3) 2 , a multiscale measure designed for the assessment of psychological and behavioral traits common in anorexia nervosa.As regards this last scale, we considered the following subscales: Drive for Thinness, Body Dissatisfaction, and Bulimia.Based on these scales, we were able to identify and exclude one participant with pathological scores.Then, their height and weight were measured to allow the calculation of the BMI.Participants' demographic information is reported in Table S1.

Behavioral measures
Food preferences Individual food preferences were assessed online by asking participants to report their liking rating on a 6point Likert scale ranging from 0 (I don't like it at all) to 6 (I extremely like it) for 100 foods (50 HC and 50 LC foods).Food stimuli were taken from the Food-pics database, a standardized and validated database of food pictures differing in caloric and macronutrient content 3 .HC food stimuli included caloric-dense and fat foods such as pizza, hamburger, and chocolate snacks, whereas LC foods mainly consisted of vegetables (e.g., spinaches, salad, carrots) and low-fat foods rich in proteins (e.g., fish, grilled chicken meat).Six of the most liked HC foods and six of the least liked LC foods were chosen as stimuli for each participant (see Table S2).

Food craving
Food craving was assessed by means of the Food Craving Questionnaire -Trait (FCQ-T) 4,5 , a self-report measure evaluating food craving in general.The questionnaire has 39 items that can lead to a total score (higher total score, higher trait food cravings) and to several subscales scores.Prior to fMRI, food craving was assessed by means of the Food Craving Questionnaire -State (FCQ-S) 4,5 , a self-report measure that evaluates the current food craving.The questionnaire has 15 items that can lead to a total score (higher total score, higher state food cravings) and to several subscales scores.

Hunger and thirst
Prior to the fMRI scanning, hunger was assessed by means of a 100-mm visual analogue scale (VAS) ranging from 0 (not hungry at all) to 100 (extremely hungry).Thirst was assessed by means of a 100-mm VAS ranging from 0 (not thirsty at all) to 100 (extremely thirsty, see Figure S1A in Supplementary Information).

Food cue-reactivity imagery paradigm
fMRI scanning was performed during a food cue-reactivity paradigm involving visually guided mental imagery.Participants were instructed to imagine eating the food (liked HC and least liked LC food) or to imagine drinking a glass of water, depending on the picture displayed on the screen.Subjects were instructed to focus on the multisensory experience of eating the food, including its smell, taste, and texture experienced in the mouth.At the end of the fMRI task, participants were asked to rate the quality of their imagery performance separately for HC, LC foods and water by means of 100-mm VAS ranging from 0 (extremely bad) to 100 (extremely good; see Figure S1A).The task included 4 blocks: (i) HC food imagery, (ii) control water imagery (after HC food), (iii) LC food imagery, (iv) control water imagery (after LC food).Each block for food (HC, LC) and water (after HC, after LC) imagery lasted 18 s, during which participants were asked to perform mental imagery while the food/water picture remained on the screen.After each block, a blank screen with the string "stop imagining" was displayed for 2 s (rest period).Trials followed a fixed order, whereby a HC imagery block (HC imagery -water imagery) was followed by a LC imagery block (LC imagery -water imagery).The HC and LC imagery blocks were repeated 6 times.The water picture represented a glass of water (W; see Figure S1B).Since data of self-report imagery quality significantly deviated from normal distribution, differences in imagery quality across the three conditions (HC, LC, W) were evaluated by means of a non-parametric repeated measures ANOVA (Friedman test).We also employed Bonferroni-corrected Wilcoxon signed-rank tests were performed to test for pairwise differences in imagery quality.Differences in imagery quality between food (average of HC and LC) and water were tested with a paired-samples t-test, as data did not deviate significantly from the normal distribution.Results were deemed significant with p < .05.The analyses were performed in R, using the "rstatix" package.

Task functional activation analyses Preprocessing
After image reconstruction, raw data visualization and conversion from DICOM to the NIFTI format were performed with MRIcron (www.mricro.com)software.All subsequent data analyses were performed in MATLAB R2016b (Mathworks.Natick.MA.USA) using the software Statistical Parametric Mapping (SPM12, Wellcome Department of Imaging Neuroscience, London, UK).Functional data were realigned using SPM realign & unwarp procedure, where all scans were coregistered to a reference image (first scan of the run) using a least squares approach and a 6 parameter (rigid body) transformation and resampled using b-spline interpolation to correct for motion and magnetic susceptibility interactions.Potential outlier scans were identified using ART as acquisitions with framewise displacement above 0.9 mm or global BOLD signal changes above 5 standard deviations.Two experimental subjects exhibited more than 20% outlier scans in the whole experimental run and were excluded from the subsequent statistical analyses, for a final sample of 25 participants.Functional and anatomical data were normalized into standard MNI space, segmented into grey matter, white matter, and CSF tissue classes, and resampled to 2 mm isotropic voxels following a direct normalization procedure using SPM unified segmentation and normalization algorithm 6,7 .Last, the stereotactically normalized scans were smoothed using a Gaussian filter of 10 x 10 x 10 mm to improve the signal-to-noise ratio, making the data suited for cluster level correction for multiple comparisons 8 .

First level analyses
Individual scans were weighted by a boxcar signal characterizing each individual experimental condition (i.e., imagine HC, imagine WHC, imagine LC, imagine WLC, rest) convolved with an SPM canonical hemodynamic response function and rectified 9 .Global differences in the fMRI signal were removed from all voxels with proportional scaling.High-pass filtering (128 s) was used to remove artefactual contributions to the fMRI signal, such as physiological noise from cardiac or respiratory cycles.A fixed-effect analysis was performed for each subject to characterize the BOLD response associated with the task before entering the relevant individual contrast images into a randomeffect analysis.At the first level, we characterized the brain activity recorded between the appearance and disappearance of the food or water's pictures, namely the entire imagery process.We included one regressor for each condition (HC food, water after HC food, LC food, water after LC food, rest) for a total of five regressors.The parameters obtained from the realignment and outlier detection procedure were added as non-interest regressors as well to partial out the impact of motion artifacts and outlier scans on the estimates of the beta parameters.
For each participant, we considered an implicit baseline leading to four contrast images: (i) HC food, (ii) LC food, (ii) water after HC food (WHC), and (iv) water after LC food (WLC).

Second level analyses
In a second-level full-factorial random-effect analysis, two contrasts were computed in order to (i) identify the neural networks commonly involved in mental imagery of food and water (food ∩ water) and (ii) distinguish the brain areas more active during the imagery of food compared to water (food > water): i. Conjunction effect of food (HC,LC) and water (WHC,WLC) imagery [FOOD ∩ WATER].The conjunction effect was tested to identify the neural networks commonly involved in mental imager of food and water.ii.
Effect of food (HC,LC) over water (WHC,WLC) imagery [FOOD > WATER].This contrast was performed to distinguish the brain areas more active during the imagery of food (HC and LC) compared to water (baseline condition).Cluster-level inferences were based on parametric statistics from Gaussian Random Field theory 10 .Results were thresholded using a combination of a voxel-level p < 0.001 uncorrected threshold, and a familywise corrected p-FWE < 0.05 cluster-size threshold across the whole brain 8 .

Task functional connectivity analyses
Preprocessing and denoising After image reconstruction, raw data visualization and conversion from DICOM to the NIFTI format were performed with MRIcron (www.mricro.com)software.Data analyses were performed in MATLAB R2016b (Mathworks.Natick.MA.USA) using the software CONN 11 and SPM12.Functional data were realigned using SPM realign & unwarp procedure, where all scans were coregistered to a reference image (first scan of the run) using a least squares approach and a 6 parameter (rigid body) transformation and resampled using b-spline interpolation to correct for motion and magnetic susceptibility interactions.Temporal misalignment between different slices of the functional data was corrected following SPM slice-timing correction procedure, using sinc temporal interpolation to resample each slice BOLD timeseries to a common mid-acquisition time.Potential outlier scans were identified using ART as acquisitions with framewise displacement above 0.9 mm or global BOLD signal changes above 5 standard deviations.Two experimental subjects exhibited more than 20% outlier scans in the whole experimental run and were excluded from the subsequent statistical analyses (the same two participants excluded in the previous task-based fMRI analysis).Functional and anatomical data were normalized into standard MNI space, segmented into grey matter, white matter, and CSF tissue classes, and resampled to 2 mm isotropic voxels following a direct normalization procedure using SPM unified segmentation and normalization algorithm 6,7 .Last, the stereotactically normalized scans were smoothed using a Gaussian filter of 10 x 10 x 10 mm to improve the signal-to-noise ratio, making the data suited for cluster-level correction for multiple comparisons 8 .In addition, functional data were denoised by including the regression of potential confounding effects characterized by white matter timeseries (5 CompCor noise components), CSF timeseries (5 CompCor noise components), motion parameters and their first order derivatives (12 factors), outlier scans, session and task effects (imagine HC, imagine WHC, imagine LC, imagine WLC, rest) and their first order derivatives, and linear trends, followed by high-pass frequency filtering of the BOLD timeseries above 0.008 Hz.CompCor 12,13 noise components within white matter and CSF were estimated by computing the average BOLD signal as well as the largest principal components orthogonal to the BOLD average, motion parameters, and outlier scans within each subject's eroded segmentation masks.

First level analyses
Seed-based functional connectivity maps were estimated characterizing the spatial pattern of functional connectivity with the seed area, as defined by the bilateral VTA of the AAL3 template 14 .Functional connectivity strength was represented by Fisher-transformed bivariate correlation coefficients from a weighted general linear model (weighted-GLM), estimated separately for each seed area and target voxel, modeling the association between their BOLD signal timeseries.Individual scans were weighted by a boxcar signal characterizing each individual experimental condition (i.e., imagine HC, imagine WHC, imagine LC, imagine WLC, rest) convolved with an SPM canonical hemodynamic response function and rectified.

Second level analyses
Group-level analyses were performed using a General Linear Model (GLM) conforming to a 2 x 2 repeatedmeasures ANOVA, with food (HC vs. LC) and water (WHC vs. WLC) as within-subjects factors.For each individual voxel a separate GLM was estimated, with first-level connectivity measures at this voxel as dependent variables (one independent sample per subject and one measurement per experimental condition), and total trait food craving scores as independent variable.Voxel-level hypotheses were evaluated using multivariate parametric statistics with random-effects across subjects and sample covariance estimation across multiple measurements.Inferences were performed at the level of individual clusters (groups of contiguous voxels).Linear regressions were performed to test whether trait food craving scores were associated with differences in whole-brain functional connectivity of the VTA during the following contrasts (food-by-water interactions): i.
[HC > WHC] > [LC > WLC].To identify the brain areas whose connectivity with the VTA during imagery of HC compared to LC foods is positively associated with trait food craving scores; ii.
[LC > WLC] > [HC > WHC].To identify the brain areas whose connectivity with the VTA during imagery of LC compared to HC foods is positively associated with trait food craving scores.Cluster-level inferences were based on parametric statistics from Gaussian Random Field theory 10 .Results were thresholded using a combination of a voxel-level p < 0.001 uncorrected threshold, and a familywise corrected p-FWE < 0.05 cluster-size threshold across the whole brain 8 .

Supplementary Table 2 Table S2. Stimuli selection for the food cue-reactivity imagery task |
The most liked high-calorie and the least liked low-calorie food stimuli selected for each participant are reported.For both low and high-calorie food, the identification number of images from the Food-pics 3 database, the calories of the food, and the liking, wanting, and frequency scores are listed.

Table 3 Table S3 | fMRI results for the task activation analyses | A.
Results of the conjunction analysis: brain regions significantly active during imagery of food stimuli and water.B. Results of the contrast analysis: brain regions significantly more active during imagery of food compared to water.For each region, we reported the anatomical labelling according to the AAL3 template 14 (Brodmann area), the coordinates in MNI space, and the Z-score.*, peak-level FWE-corrected.