For healthy controls, inclusion criteria were (1) aged 40–75 years, (2) provision of written informed consent, (3) normal or corrected-to-normal vision, and additionally for patients, (4) a diagnosis of PD following UK Brain Bank criteria. Exclusion criteria for all participants were (1) current use of psychotropic medication other than levodopa, dopamine-agonists or other Parkinson-medication, (2) major somatic disorder, (3) current psychiatric diagnosis as established by a psychiatrist, (4) presence of dementia, history of stroke or other neurological diseases (as established by neurologist). An additional screening for dementia was performed using the montreal cognitive assessment (MoCA), a screening tool for cognitive dysfunction (Nasreddine et al. 2005), with a cutoff for dementia according to Biundo and colleagues (2014). Patients were recruited through outpatient clinics. Healthy controls were recruited through advertisement in local newspapers, online advertisement and through participating patients (e.g., spouses, relatives, etc.).
The study was approved by the medical ethical committee of the VU Medical Center, Amsterdam. Informed consent was obtained from all individual participants included in the study. All methods were carried out accordance with relevant guidelines and regulations.
This study was part of a larger cross-sectional case–control study investigating visual attention, reward and pain in PD. Study size was based on expected learning and attention differences between groups. Note that only procedures concerning this project will be described. Patients visited the hospital twice: on the first visit, the MoCA was administered, and the questionnaires were handed in (filled out just prior to visit). During the second and third visit, the MRI was performed in either the ON or OFF phase. The same procedure was completed for the controls, with the exception that they underwent only one MRI session. The MRI was planned in the same week as the clinical assessment in almost all patients, but as a rule no later than 60 days after the clinical assessment. For the MRI, patients were invited to the hospital in the afternoon for the ON phase, and in the morning for the OFF phase. ON and OFF phase as the first or second MRI session was counter-balanced across participants. The OFF phase was defined as at least 12 h of dopaminergic medication overnight withdrawal. One patient took their medication 8.5 h before the resting-state scan to relieve symptoms.
Here, we define clinical pain as naturally occurring pain that is not experimentally induced, regardless of origin or type. Ultimately, all pain processing is a neurophysiological phenomenon, irrespective of origin (Garland 2012). Pain was measured during the ON and OFF phase by means of the colored analogue scale (CAS) for intensity as well as for pain affect: subjects were asked to indicate the intensity of their pain (CAS intensity) and how much they were bothered by their pain (CAS affect) both on a scale ranging from ‘None’ (light pink, 0) to ‘Maximal’ (dark red, 100) (McGrath et al. 1996). The Dutch version of the McGill Pain Questionnaire was administered during the first visit to inquire about pain during the previous month (Melzack 1975). We utilized the total score on the number of words chosen (NWC) part of the McGill pain questionnaire. The NWC consists of three major classes of pain descriptors, which were used by the subjects to specify their pain experience. These classes are of sensory, affective or evaluative nature (van der Kloot et al. 1995). Total score on the NWC was used as score for each subject’s clinical pain experience.
Imaging data were collected with a 3T GE Signa HDxT (General Electric, Milwaukee, WI, USA) at the VU University Medical Center (Amsterdam, The Netherlands). Structural images were acquired with a 3D T1-weighted MP RAGE sequence with the following acquisition parameters: voxel size = 1 mm isotropic, 176 slices, 256 × 256 matrix, repetition time (TR) = 8.2 ms, echo time (TE) = 3.2 ms, flip angle (FA) = 12°, inversion time (TI) = 450 ms. Resting-state data were acquired using a T2*-weighted echo-planar functional scan: number of volumes = 202, 42 slices, slice thickness = 3.2 mm, matrix size = 64 × 64, TR = 2150 ms, TE = 35 ms, FA = 80°, field of view = 240 mm, total duration 7:12 min, voxel size was 3 mm with 0.3 mm spacing. For the resting-state scan, subjects were instructed to close their eyes, lie still and avoid falling asleep. Participants’ heads were immobilized using foam pads to reduce motion artifacts.
Processing of fMRI data
Data were analyzed using FSL FMRIB software library v5.0.9 (Jenkinson et al. 2012) and custom-built scripts in bash and Matlab, version 2015a (Mathworks, Natick, MA, USA). The following pre-processing steps were taken: (1) images were corrected for head motion (using MCFLIRT), (2) slice-timing correction was applied, (3) non-brain tissue was removed (using Brain Extraction Tool, BET), (4) functional images were registered to subject-space (T1-weighted structural image) using BBR, (5) this image was registered to MNI152 standard space (FLIRT for linear registration with 12 DOF), (6) high-pass filtering above 0.01 Hz was applied, (7) spatial smoothing was performed at 5 mm full-width half maximum (FWHM), (8) segmentation of gray and white matter was performed using FAST and SIENAX, (9) the first three volumes of each resting-state scan were discarded to achieve field equilibrium, (10) average motion was calculated as the mean of the absolute head movement over all time series for 6 DOF per individual (three translations and three rotations).
After preprocessing, a resting-state adjacency matrix (representing an undirected weighted network) was reconstructed per subject as follows. First, time series were scrubbed for motion outliers: time points with frame-to-frame displacement > 1.5 mm (6 DOF) were excluded from further analyses. The remaining time series of chosen atlas regions (see below) were used to calculate the connectivity matrix using Pearson’s correlation coefficients between time series of each pair of regions included. A Fisher transformation on these correlation coefficients was used to get normally distributed correlation values. Atlas regions were based on the atlas of Power et al. (2011). The default mode network (DMN) was constructed from 58 of these nodes (see Power et al. 2011 for all atlas regions and specification of the DMN). To form a pain network, we additionally used a subdivision of the NPS of Wager et al. (2013). In their paper, Wager and colleagues based their NPS on 32 areas (Wager et al. 2013). Sixteen of these areas had positive predictive weights for physical pain, and the other 16 had negative predictive weight. For the current study, only the areas of the NPS with positive predictive weights were used to form a pain network because the interpretation of a network of positive predictive weights is most straightforward, particularly when making comparisons between patient/control groups and medication sessions. See Fig. 1 for a visual overview of this 16-node pain network.
The resting-state adjacency matrix was used to calculate network topology. We calculated betweenness centrality (BC) and global efficiency (GE) of the resting-state adjacency matrix. BC measures how central a node lies with respect to the rest of the network, and is based on how many shortest paths pass through it. Nodes with high BC represent hubs of the network (Fornito et al. 2016). GE represents efficiency of the underlying network and is a measure of integration. The Brain Connectivity Toolbox (brain-connectivity-toolbox.net) was used to calculate GE per network, and BC of all nodes in the network (with Matlab scripts: efficiency_wei.m and betweenness_wei.m, respectively) (Rubinov and Sporns 2010). BC was then normalized (z-scored) per participant, and averaged per network. GE and average BC for all 16 pain nodes represented the pain network, GE and BC for the 58 DMN nodes represented the DMN. For the whole brain (264 nodes), only GE was calculated, since calculating BC (or ‘hubness’) of the whole brain is essentially meaningless.
Mann–Whitney U test was performed to investigate the difference between PD and controls on all pain measures, as they were not normally distributed. To investigate the differences in pain and network topology, the independent variable ‘group’ (patients ON and OFF, and controls) and dependent variables BC and GE measures, were entered in multivariate analyses of covariance (MANCOVAs), with average motion during the scan as a covariate. Only differences between patients and controls were considered. To investigate the relationship between pain and network topology, a hierarchical stepwise linear regression was performed per group (controls, PD ON, PD OFF), with the BC and GE of the pain network as independent variables, and clinical pain as dependent variable. In each analysis’ first block, average movement during the scan was added as a covariate to account for motion in the scanner, and a forward stepwise method was utilized to investigate the contribution of each separate independent variable. Scores on the NWC served as a dependent variable. The alpha level was set at 0.05, and tested two-sided.
The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.