https://doi.org/10.5061/dryad.1vhhmgr2r

Code/Software

We include the code for running pre-processing and analyses:

1. Preprocessing of fMRI and fPET datasets (see folder 01_Preprocessing / matlab code, requires SPM12).
2. Univariate 1st level analysis (see folder 02_1st_Level_fMRI_GLM_analysis / matlab code, requires SPM12).
3. Connectivity analysis (see folder 03_Connectivity / matlab code, requires SPM12).
4. Multivariate analysis (see folder 04_MVPA / Python code)

Description

1. Preprocessing of fMRI and fPET datasets.
   Data preprocessing was conducted as using Statistical Parametric Mapping 12 (SPM 12, Wellcome Trust Centre for Neuroimaging, University College London, London, United Kingdom) via Matlab (The MathWorks, Natick, MA, USA) and Analysis of Functional NeuroImages (AFNI, National Institute of Mental Health (NIMH), Bethesda, Maryland, USA). In summary, realignment of fMRI and fPET data was performed in SPM. Binary masks were generated from average images and the anatomical MRI scans. With these, the brain was extracted from the fPET, anatomical reference and fMRI image (“skull-stripping”) before co-registration of the fPET and fMRI to the anatomy. Spatial normalization was performed using parameters, which were calculated by comparing the anatomical reference to the Schiffer rat brain atlas. The normalized fMRI and fPET images were smoothed using a 1.5 × 1.5 × 1.5 mm3 Gaussian kernel towards the spatial resolution of the PET insert. A temporal high-pass filter with a cut-off frequency of 256 Hz was applied to the fMRI data, with the purpose of removing scanner attributable low frequency drifts in the fMRI time series.
2. Univariate 1st level analysis (fMRI statistical analysis).
   Data were analyzed using Statistical Parametric Mapping (SPM), version 12 (www.fil.ion.ucl.ac.uk/spm). A block design was employed for the ChR2 and the GFP groups modeling each of the six 10-minute stimulation blocks using a canonical hemodynamic response function that emulates the early peak at 5 seconds and the subsequent undershoot (89). The within-subject design matrix for the first level analysis included two regressors: optogenetic stimulation (OGS) and baseline (3 minutes between stimulation blocks). Two contrast images per individual were calculated: OGS > baseline and baseline > OGS.
3. Connectivity analysis.
   We performed seed-based connectivity analysis, on both the fMRI and fPET datasets, to evaluate connectivity differences between optogenetic stimulation and baseline blocks. The striatum receives axonal projections from SN dopaminergic neurons and houses a large population of GABAergic neurons. Consequently, we used the right striatum as seed region and assessed connectivity differences in response to the optogenetic stimulation.
4. Multivariate analysis
   For conducting MVPA ROI analysis, we performed an initial feature selection by considering only those voxels within the right SN. Estimated responses across relevant voxels from the right SN formed the feature vectors used to train the classifier. We evaluated a Gaussian Naive Bayes classifier (GNB) on the fMRI and fPET datasets. Analogous to the searchlight analysis, to assess multivariate effects, we used an across-subject classification approach. A positive result in this framework implies that the model has learnt an implicit rule from the training data that yields statistically significant generalization power on data from new subjects. We employed a leave-one-subject-out cross-validation for the MVPA ROI analysis. We tested the statistical significance of the accuracy of prediction using a permutation test: assuming there is no class information in the data, the labels can be permuted without altering the expected accuracy using a given classifier and number of features (i.e., this would equal chance level) (27). We performed 1000 permutations of a leave-one-subject-out cross-validated MVPA, and evaluated a GNB classifier with default parameters in Scikit-learn.

Description of data sets:

Four data sets:

Example: swrrFUNC_ChR2-003

BOLD-fMRI data (preprocessed): swrrFUNC
[18F]FDG-PET data (preprocessed): swrrrPET
GFP = control animals, injected with an adeno-associated viral vector (AAV) to overexpress GFP
ChR2 = animals, injected with an AAV to overexpress ChR2
number = animal ID