YAEH Controlled-Feeding Study: Neuroimaging Data and fMRI Analysis

1. Study overview

This OpenNeuro dataset contains available de-identified, defaced neuroimaging data from a randomized, controlled crossover feeding trial investigating the effects of diets high in versus free of ultra-processed foods (UPFs) on brain responses to a milkshake cue-and-consumption task in people aged 18–25 years.

Participants completed two 14-day, eucaloric, nutritionally matched diet periods separated by a four-week washout: a UPF diet providing approximately 81% of energy from UPFs and a non-UPF diet providing 0% of energy from UPFs. Functional MRI was collected before and after each dietary condition. The dataset includes available scans beyond the final sample used in the corresponding publication.

• Clinical trial: NCT05550818
• Study publication: Effects of Controlled Diets High in and Free of Ultraprocessed Food on the Brain of Emerging Adults, accepted in Obesity. [Add the final journal DOI/citation when available.]

2. BIDS organization and task files

Data are organized by participant (sub-YAEH##), session, and imaging modality. Session labels correspond to the four study time points:

Session	Meaning
ses-preUPF	Before the UPF intervention
ses-postUPF	After the UPF intervention
ses-preMPF	Before the non-UPF intervention
ses-postMPF	After the non-UPF intervention

The functional files labeled task-taste1 and task-taste2 are two runs of the same milkshake cue-and-consumption task, not two different task paradigms. Each run contained milkshake and tasteless conditions with separate cue and consumption/delivery events. Not every participant has every session or run.

BIDS-compatible *_events.tsv files are included for the available task runs with converted event information. These contain the stimulus timing information used to specify the task conditions:

• onset: event onset in seconds, relative to the corresponding BOLD run;
• duration: event duration in seconds;
• trial_type: task condition name taken from the original SPM task-regressor file.

The events tables were generated from the original SPM multiple-condition MATLAB (.mat) task-regressor files using R and the R.matlab package. Original task-regressor files are retained under sourcedata/, organized by subject and session. The resulting .tsv files are stored alongside their matching BOLD images in each participant's func/ folder. The final repository uses only the standardized BIDS session labels described above.

Example organization (illustrative filenames; file availability varies):

README.md
dataset_description.json
participants.tsv
sourcedata/
  sub-YAEH01/
    ses-preMPF/
      func/
        task_regressor_taste1.mat
        task_regressor_taste2.mat
sub-YAEH01/
  ses-preMPF/
    anat/
      ...
    fmap/
      ...
    func/
      sub-YAEH01_ses-preMPF_task-taste1_bold.nii.gz
      sub-YAEH01_ses-preMPF_task-taste1_bold.json
      sub-YAEH01_ses-preMPF_task-taste1_events.tsv
      sub-YAEH01_ses-preMPF_task-taste2_bold.nii.gz
      sub-YAEH01_ses-preMPF_task-taste2_bold.json
      sub-YAEH01_ses-preMPF_task-taste2_events.tsv

Event-file scope: The presence of task events for a given run depends on availability of the corresponding source timing file and BOLD scan. Event files derived from SPM design inputs document the conditions used for modeling; they should not be assumed to contain every field available in the original experimental stimulus log. For secondary analyses, users should check the timing, run alignment, and included condition definitions.

3. Participants and analysis inclusion

The parent study enrolled 36 participants, of whom 28 completed the study. All available de-identified, defaced neuroimaging scans are shared, including available data for participants who did not complete the study or were not included in the published fMRI analyses. Inclusion in this deposit does not imply inclusion in the analytical sample.

Did not complete the study (n = 8): YAEH05, YAEH07, YAEH08, YAEH14, YAEH19, YAEH21, YAEH28, YAEH33.

Completed the study but were excluded from the final fMRI analysis (n = 6): YAEH01, YAEH09, YAEH17, YAEH18, YAEH23, YAEH36. Among study completers, reasons for exclusion were missing data resulting from a power outage (n = 1), equipment malfunction (n = 2), illness during testing (n = 1), and imaging quality-control criteria (n = 2). The individual reasons are not linked to participant identifiers here.

Included in the published fMRI analysis (n = 22): YAEH02, YAEH03, YAEH04, YAEH06, YAEH10, YAEH11, YAEH12, YAEH13, YAEH15, YAEH16, YAEH20, YAEH22, YAEH24, YAEH25, YAEH26, YAEH27, YAEH29, YAEH30, YAEH31, YAEH32, YAEH34, YAEH35.

The final fMRI sample included 7 late adolescents (18–21 years) and 15 young adults (22–25 years). Analyses required all four imaging sessions, with at least one usable functional run per session. Run-level quality criteria included excluding runs with motion exceeding 2 mm in a single direction or more than 20% of volumes flagged for excessive DVARS (greater than 1.5 standard deviations above the mean).

4. MRI acquisition and task design

Images were collected on a Siemens Magnetom Prisma 3T scanner with a 64-channel head coil. Each study session included structural imaging, phase-encoding reference images for susceptibility-distortion correction, and up to two approximately 11-minute task fMRI runs. The BOLD acquisition used an EPI sequence (TR = 1,500 ms; TE = 34 ms; 2-mm isotropic voxels; 72 slices; multiband factor 4). Reverse phase-encoding data were acquired for distortion estimation.

During the task, participants viewed cues predicting milkshake or tasteless solution and received the corresponding liquid stimulus. Trials included both cue and delivery/consumption phases. Timing included jittered intervals to accommodate swallowing and improve estimation of the event-related responses. Task presentation and pump triggering were controlled using PsychoPy. See the publication for full stimulus, trial, and imaging parameters.

5. fMRI preprocessing

Preprocessing and modeling for the published analyses were conducted using FSL and SPM12, with subsequent permutation-based inference using PALM and follow-up statistical analyses in R. The following describes the analysis workflow applied to the published fMRI sample; the OpenNeuro release primarily contains the shared BIDS imaging and event data, not a fully executed preprocessing or voxelwise-analysis pipeline.

1. Susceptibility-distortion correction: Opposite phase-encoding images were used with FSL topup to estimate and correct EPI susceptibility-related distortion/unwarping.
2. Motion correction and alignment: Functional images were realigned in SPM12 and co-registered with participant-specific structural images.
3. Spatial normalization: Images were normalized to standard stereotaxic space in SPM12.
4. Spatial smoothing: Functional images were smoothed with a 6-mm full-width at half-maximum Gaussian kernel.
5. Motion/artifact assessment: Functional runs were screened for excessive motion and DVARS outliers. Motion and artifact-related nuisance regressors were included in first-level modeling, as described below.

Details and implementation choices are provided in the associated manuscript. The distributed raw BIDS images should not be confused with the preprocessed images or contrast maps used for inference.

6. First-level (within-session) statistical models

First-level general linear models were fitted in SPM12, separately for each participant/session, using the two available runs as applicable. The event-related design included distinct regressors for:

• Milkshake cue;
• Tasteless cue;
• Milkshake consumption/delivery;
• Tasteless consumption/delivery.

Task events were specified using their recorded onsets and durations, represented with boxcar regressors convolved with the hemodynamic response function; temporal derivatives were included. Nuisance modeling accounted for motion-related effects and non-task signals, including movement estimates/expanded motion terms, framewise displacement (FD), DVARS-related outlier volumes, white matter/cerebrospinal fluid signals, and rinse events as applicable to the model.

The principal first-level contrasts compared milkshake versus tasteless for each task phase:

• Cue contrast: milkshake cue > tasteless cue;
• Delivery contrast: milkshake consumption > tasteless consumption.

Reverse contrasts (tasteless > milkshake) were also examined. The resulting participant/session contrast images were taken forward to group-level analyses. In the original analysis workflow, the primary milkshake-versus-tasteless cue and delivery contrasts were designated con_0001.nii and con_0002.nii, respectively.

7. Group-level parametric analyses

Second-level models in SPM12 tested the effects of dietary intervention on BOLD response separately for the cue and delivery contrasts. The primary repeated-measures factors were:

• Diet: UPF versus non-UPF;
• Time: before versus after each diet.

The primary comparison was the Diet × Time interaction, testing whether pre-to-post changes in response differed between dietary conditions. Separate exploratory models assessed Diet × Time × Age Group, as well as interactions involving sex and BMI. Age groups were defined as 18–21 and 22–25 years.

Region-of-interest (ROI) analyses emphasized reward- and cognitive-control-related areas, including prefrontal and striatal regions. Small-volume/ROI correction was used when appropriate; statistical significance for reported findings was evaluated using family-wise error correction (pFWE < 0.05). Whole-brain results were also evaluated with FWE correction. More liberal thresholds (e.g., uncorrected p < 0.001 and cluster extent > 10 voxels) were used for visualization and were not substituted for corrected statistical significance.

8. Nonparametric permutation analysis (PALM)

A complementary analysis used PALM (Permutation Analysis of Linear Models) with 10,000 permutations to evaluate the Diet × Time × Age Group effect without relying solely on parametric inference.

For each participant with all four sessions, the within-diet pre-to-post change was calculated from the first-level contrast images:

UPF change       = postUPF - preUPF
Non-UPF change   = postMPF - preMPF
Difference score = (postUPF - preUPF) - (postMPF - preMPF)

This produced one difference-of-differences image per participant for the relevant contrast. Images were concatenated into a 4D image and compared between the two age groups (18–21 versus 22–25 years), with separate variance groups to accommodate unequal group variances. Contrasts evaluated both directions of the between-group difference. Permutation testing used the designated prefrontal cortex/striatal ROI mask and controlled family-wise error across the tested voxelwise comparisons/age-group contrast directions as specified in the published analysis.

A positive difference score indicates a larger pre-to-post change under the UPF condition than under the non-UPF condition; a negative score indicates the reverse. Cue and delivery contrasts were processed separately.

9. Follow-up and brain–behavior analyses

Follow-up analyses were conducted in R to characterize significant interactions and associations:

• Post hoc condition comparisons: Estimated marginal means and planned/pairwise comparisons were evaluated with emmeans, using Tukey adjustment where applicable.
• Permutation-based follow-up tests: Additional nonparametric analyses used permuco with 10,000 permutations.
• Continuous age: Participant-level difference-of-differences estimates were examined as a function of age to complement the categorical age-group analysis.
• Behavioral associations: Relationships between neural responses and subsequent ad libitum buffet energy intake, eating in the absence of hunger, and habitual UPF intake were investigated using the reported regression and covariate models. Habitual UPF consumption was derived from dietary recalls and expressed as the percentage of food-and-beverage grams classified as UPF.

Parameter estimates from identified regions were extracted for visualization of the published effects. Extracted parameter-estimate data, R figure scripts, and further pipeline documentation are intended to be archived separately in VTechData. [Add the VTechData dataset URL or DOI when available.] These supplementary materials reproduce figure-level visualizations but are not substitutes for the complete first- and second-level voxelwise models.

10. Scope, reuse, and interpretation

• The repository includes available defaced imaging and corresponding standardized task-event files. The number of available participants, sessions, and runs may differ from the final published sample.
• Original SPM task-regressor .mat files may be retained in sourcedata/ for provenance; they are not BIDS-native events tables.
• This dataset does not by itself contain every intermediate SPM/FSL/PALM output, voxelwise statistical map, first-level design matrix, or analysis script used in the publication.
• Reproducing the published statistical results requires applying the documented preprocessing, quality-control criteria, modeling choices, ROI definitions, and participant selection to the shared data; reproduction may also require inputs available through the separate VTechData deposit.
• Consult the publication and the BIDS sidecars for the full acquisition details and task definitions. Secondary users should verify event timing relative to the stored BOLD volumes and inspect any available run-specific omissions before modeling.

11. References and links

• Study: Effects of Controlled Diets High in and Free of Ultraprocessed Food on the Brain of Emerging Adults. Accepted in Obesity. [Insert final DOI.]
• Trial registration: https://clinicaltrials.gov/study/NCT05550818
• BIDS specification: https://bids-specification.readthedocs.io/
• Related analysis data and code: VTechData link to be added when available.