Project Information and Reproducibility Guide

View the Project on GitHub PennLINC/ZAPR01_dMRI_TMSfMRI


Project Lead

Valerie J. Sydnor

Faculty Leads

Desmond J. Oathes
Theodore D. Satterthwaite

Analytic Replicator

Matthew Cieslak


Romain Duprat, Hannah Long, Matthew W. Flounders, Joseph Deluisi, Morgan Scully, Nicholas L. Balderston, Yvette I. Sheline, Dani S. Bassett

Project Start Date

August 2018

Current Project Status

Manuscript in Submission


ZAPR01-Healthy Controls

Github Repository


Path to Data on Filesystem Dopamine


$ZAP/Quality_Control : Study QC spreadsheet
$ZAP/qsiprep_0.6.3 : QSIPrep output
$ZAP/mrtrix_fixelanalysis : output of all mrtrix-based analyses $ZAP/sites_of_stim : TMS SOS coordinates, ROIs, masks
$ZAP/templates : templates and ROI masks
$ZAP/output_measures : spreadsheets with study measures for stats

Conference Presentations

Poster presentation at The Society of Biological Psychiatry Annual Meeting, April 2021. Amygdala TMS-fMRI Evoked Response is Influenced by Prefrontal-Amygdala White Matter Pathway Fiber Density


The analytic workflow implemented in this project is described in detail in the following sections. Analysis steps are described in the order they were implemented; the script(s) used for each step are identified and links to the code on github are provided.

Diffusion MRI Preprocessing and Reconstruction

ZAPR01 diffusion MRI data was first BIDSifyed, preprocessed with QSIPrep, and postprocessed with mrtrix’s fixel-based analysis pipeline, as detailed below:

  1. Organize Data into BIDS
    • ZAPR01 data was organized into the Brain Imaging Data Structure via heudiconv-0.5.4 via the script /bids/ZAP2BIDS.sh. This script both converts the original ZAPR01 directory/file structure to BIDS and adds IntendedFor specifications to fieldmap jsons
    • The study-specific heuristic /bids/ZAPR01_legacyandnew_heuristic.py was used for BIDS conversion via the call
      $ sh ZAP2BIDS.sh vsydnor /data/jux/oathes_group/projects/vsydnor/scripts/ZAP2BIDS/ZAPR01_legacyandnew_heuristic.py
    • Note: heudiconv was run on chead (BIDS output on chead: /data/jux/oathes_group/studies/TNI_ZAPR01/MRI/BIDS)
  2. Preprocess Diffusion Data with QSIPrep
    • ZAPR01 single shell diffusion data were preprocessed with qsiprep version 0.6.3RC3
    • First, the singularity image qsiprep-0.6.3RC3.simg was built from docker by executing /qsiprep/build_qsiprep_0.6.3.simg.sh
    • QSIPrep was then run via a job array by executing /qsiprep/run_qsiprep0.6.3_ZAPR01_DTI.sh. QSIPrep was run with the following parameters:
      $ qsiprep-0.6.3RC3.simg --bids_dir $bidsdir --output_dir $qsiprepdir --analysis_level participant --participant_label $SUBJ --hmc_model eddy --eddy-config eddy_params.json --b0-motion-corr-to first --output-space T1w --output-resolution 1.3 --force-spatial-normalization --do-reconall
    • Note: qsiprep-0.6.3RC3.simg was built and run on chead, and the output of qsiprep was copied to dopamine (/storage/vsydnor/ZAPR01_WhiteMatter_TMSfMRI/qsiprep_0.6.3). All subsequent scripts/analyses were run on dopamine
  3. Postprocess Diffusion Data with MrTrix Fixel-Based Analysis Pipeline
    The mrtrix fixel-based analysis pipeline was implemented to postprocess the output of qsiprep. This pipeline includes diffusion signal reconstruction via constrained spherical deconvolution, fixel segmentation, and tractography generation. The pipeline generates subject-specific FOD images and a study-specific FOD template, subject-specific fixel images and a study-specific fixel template, and template-based whole-brain tractography. The pipeline outlined below includes the relevant steps documented in mrtrix3Tissue and mrtrix CSD fixel pipeline and was executed as follows:

TMS Sites of Stimulation – Left Amygdala Tract Extraction

Following preprocessing, FOD/fixel reconstruction, and tractography, the whole-brain tractography was used to delineate a putative causal pathway by which TMS-evoked cortical activity could travel to the left amygdala. To identify this pathway, a study-specific TMS cortical sites of stimulation mask was generated, and streamlines with endpoints in this mask and in the left amygdala were extracted, as follows:

  1. Generate a Cortical Sites of Stimulation Mask for Amygdala-targeted TMS
    • For each subject, the TMS site of stimulation for amygdala-targeted TMS was localized to a set of X, Y, Z MNI space voxel coordinates using Brainsight NeuroNavigation information. A site of stimulation spherical ROI was generated from the MNI coordinates, and ROIs from all subjects were merged into a single TMS sites of stimulation mask. This was accomplished with the script /sites_of_stim/siteofstim_processing.sh
  2. Register Tractography Inclusion Masks from MNI Space to FOD Template To identify tractography streamlines with endpoints in the amygdala TMS sites of stimulation mask and the left amygdala, the sites of stimulation mask (generated in step 1) and a left amygdala ROI (extracted from the Harvard Oxford subcortical atlas) were transformed from MNI space to the study-specific FOD template
  3. Extract Causal Pathway Streamlines
  4. Map Pathway Streamlines to Fixels
    • In order to quantify mean fixel-based pathway measures for each subject (e.g. fiber density), the causal white matter pathway streamlines were mapped back to template fixels with /tract_analyses/tracts_to_fixels.sh
    • Tract fixels were then cropped from the whole-brain fixed mask via /tract_analyses/fixelcrop.sh (using a primary streamline threshold of 5 as well as additional streamline thresholds for sensitivity analyses)

Tract Anatomy and Brodmann Overlap

The scripts in /tract_analyses/anatomy were written to gain insight into the anatomy of the identified vlPFC-amygdala pathway.

Tract-Based Mean Fixel Measures and TMS-fMRI Evoked Response Measures

To assess whether microstructural (FD) and macrostructural (logFC) properties of the vlPFC-amygdala white matter pathway were associated with the magnitude of TMS-evoked functional response in the amydala (and to conduct related sensitivity and specificity analyses), vlPFC-amygdala white matter pathway fixel measures (FD, logFC, FDC) were extracted for each subject with /tract_analyses/fixelmeasures.sh and TMS-fMRI functional evoked response data were extracted with /TMSfMRI_EvokedResponse/TMSfMRI_SignalChange_TMSon_Measures.sh. Note: the fMRI data were preprocessed with fMRIPrep and postprocessed with XCP to generate single pulse TMS-fMRI BOLD signal change (i.e. evoked response) maps, as detailed in Duprat et al. (In preparation).

Statistical Analysis

Manuscript statistics were conducted in R and are included in /statistics/dMRI_TMSfMRI_statistics.Rmd.


Manuscript figures were generated using R, mrview, fsleyes, and Slicer. Slicer was used to visualize TMS sites of stimulation (Figure 2A, Figure 4C); the center of gravity for all vlPFC stimulation sites / control stimulation sites was calculated with /sites_of_stim/stimsites_vlPFC_control_COG.sh. Mrview was used to visualize the vlPFC-amygdala pathway, as well as pathway streamlines, FODs, and fixels (Figure 3A). Fsleyes was used to visualize pathway anatomy overlaid on the JHU white matter atlas (Figure 3B). The R code in /visualization/dMRI_TMSfMRI_visualization.Rmd was written to generate all graphs (Figure 2B, Figure 2C, Figure 4A, Figure 4B, Figure 4D).