Supplement A : fMRI data acquisition and processing Data acquisition

1 Supplement A: fMRI data acquisition and processing Data acquisition Images were acquired with a 3T Siemens TIM TRIO whole-body scanner (Siemens Symphony, Erlangen, Germany) with a 32-channel head coil. First, a high-resolution scan was acquired for anatomical referencing using a T1-weighted MPRAGE sequence (FoV: 256 mm, slice thickness 1.0 mm, TR 2300 ms, flip angle 9°, resolution:1x1x1 mm). Functional images were obtained in two sessions with a short pause in between. Concerning the functional imaging, a total of 552 volumes were acquired using a T2*-weighted gradient echo EPI with 36 slices (slice thickness=3 mm, descending slice order, TR=2.25 s, TE=30 ms, flip angle =70°, FoV=192 mm). The first 6 volumes of each functional session were discarded due to saturation effects, leaving a total of 540 volumes. Stimuli were presented with the E-Prime 2.0 presentation software1 as an event-related design in a pseudo-randomized order in 2 sessions (20 pictures of each category in each session). The pictures were shown for 4 seconds each, separated by a fixation cross; the inter-stimulus interval was 2 seconds. The resulting DICOM files were converted to 4D-NIfTI-files with the tools MRIConvert2 and dcm2nii3.

distortions using the fieldmap of each participant, and slice time corrected.
The high-resolution structural T1-weighted image of each participant was processed and normalized with the CAT12 toolbox 5 using default settings. Each structural image was segmented into gray matter, white matter, and CSF, and denoised, then warped into MNI space by registration to the DARTEL template provided by the CAT12 toolbox via the high-dimensional DARTEL registration algorithm [1]. Based on these steps, a skull stripped version of each image in native space was created.
To normalize functional images into MNI space, the functional images were coregistered to the skull stripped structural image and the parameters from the DARTEL registration were used to warp the Multi-level assessment of obsessive-compulsive disorder (OCD) reveals relations between neural and neurochemical levels https://doi.org/10.1186/s12888-020-02913-5 K. Viol et al.
BMC Psychiatry 2020 2 functional images, which were resampled to 3 × 3 × 3 mm voxels and smoothed with a 6 mm FWHM Gaussian kernel. The quality of the preprocessing was checked using the tools BXH 6 and tsdiffana 7 .

Statistical Analysis
Since SPM uses a mass-univariate approach, the effect of the conditions were modeled for each voxel with a general linear model [2]. The movement parameters gained from the realignment procedure during preprocessing were used as regressors. Freyer et al. [6] found a negative correlation of the intensity of OCD symptoms with the activity of the pallidum, Huyser et al. [7] a positive correlation with the prefrontal cortex, and Verfaillie et al. [8] reported positive correlations with left putamen, left OFC, left DLPFC, and right insula. Also, the predictive value of correlations between pre-treatment neural activity and changes in symptom severity yielded diverging results: while Sanematsu et al. [9] found predictive power for the cerebellum and the superior temporal gyrus, Olatunji et al. [10] reported that activation in the prefrontal cortex, the temporal pole, and the amygdala, was associated with better treatment response. neurochemical levels https://doi.org/10.1186/s12888-020-02913-5 K. Viol et al. BMC Psychiatry 2020 3 Multi-level assessment of obsessive-compulsive disorder (OCD) reveals relations between neural and neurochemical levels https://doi.org/10.1186/s12888-020-02913-5 K. Viol et al. BMC

Supplement C: Results of the neurochemical parameters
The changes of the neurochemical parameters, which were assessed from the serum of the patients taken at 8 a.m., did not show significantly alterations during the process of psychotherapy. The nonsignificance can probably be partly explained by the high variability of the parameters, as can be seen in the high standard deviations of Table S2.

Results of the stepwise linear regressions
The result of the stepwise bilinear regression in MATLAB for the model explaining the change in neural activity of the putamen is given in Table S3. The model's statistical characteristics were R² =.78, F(6,10)=5.93, and p = .007, but some of the variables/interaction terms were not significantly different from zero (last column in Table S3). Only the significant variables were then entered into a conventional linear regression model. Multi-level assessment of obsessive-compulsive disorder (OCD) reveals relations between neural and neurochemical levels https://doi.org/10.1186/s12888-020-02913-5 K. Viol et al. Note that none of the variables were correlated.
Two values were determined as a possible outlier ( Figure S1-A). Even though the distribution of the residuals did not suggest a distortion of this outlier ( Figure S1-B), the regression was repeated with a robust regression algorithm. The result ( Figure S1-C) was hardly distinguishable from the original regression and still significant (F(13,3) = 5.08, p = .02). The goodness-of-fit was only slightly reduced (R² = .54 compared to R² = .58). All parameters remained significant (Table S4). Multi-level assessment of obsessive-compulsive disorder (OCD) reveals relations between neural and neurochemical levels https://doi.org/10.1186/s12888-020-02913-5 K. Viol et al.