Functional connectivity in dementia with Lewy bodies: A within‐ and between‐network analysis

Abstract Dementia with Lewy bodies (DLB) is a common form of dementia and is characterized by cognitive fluctuations, visual hallucinations, and Parkinsonism. The phenotypic expression of the disease may, in part, relate to alterations in functional connectivity within and between brain networks. This resting‐state study sought to clarify this in DLB, how networks differed from Alzheimer's disease (AD), and whether they were related to clinical symptoms in DLB. Resting‐state networks were estimated using independent component analysis. We investigated functional connectivity changes in 31 DLB patients compared to 31 healthy controls and a disease comparator group of 29 AD patients using dual regression and FSLNets. Within‐network connectivity was generally decreased in DLB compared to controls, mainly in motor, temporal, and frontal networks. Between‐network connectivity was mainly intact; only the connection between a frontal and a temporal network showed increased connectivity in DLB. Differences between AD and DLB were subtle and we did not find any significant correlations with the severity of clinical symptoms in DLB. This study emphasizes the importance of reduced connectivity within motor, frontal, and temporal networks in DLB with relative sparing of the default mode network. The lack of significant correlations between connectivity measures and clinical scores indicates that the observed reduced connectivity within these networks might be related to the presence, but not to the severity of motor and cognitive impairment in DLB patients. Furthermore, our results suggest that AD and DLB may show more similarities than differences in patients with mild disease.


Independent healthy control group
from two previous studies were selected. They were significantly younger than the HCs from the main analysis, but matched in terms of overall cognition (Supplementary Table S1).
All participants were scanned on the same scanner as the participants from the main analysis.
Eighteen of the additional HC participants were scanned with a slightly different scanner protocol with a change in the TR to 2072 ms and a change in the voxel size of the resting state scans to 3 x 3 x 4 mm 3 .
The resting state data were preprocessed in the same way as described in the main manuscript. Two subjects were excluded because they exceeded the motion exclusion criteria resulting in 42 independent HC participants that were included in the generation of the RSN templates.   Although decreased connectivity in the DLB group is reported for all clusters in panels A-F of Figure   2, it was evident that some of these results were due to correlations shifting from positive in the control group to negative in the DLB group (e.g. TN-1). Similarly, increased connectivity in the DLB group could also be due to correlations being negative in HC, and shifting to positive correlations in DLB (e.g. ISN2-1), see Supplementary Figure S2.

Locations of RSN spatial maps
Anticorrelations are not easy to interpret and shifts from positive to negative or from negative to positive correlations in patient groups are even harder to understand. However, while it has previously been believed that anticorrelations might be a mere result of certain preprocessing steps [Murphy et al., 2009], it has recently been argued that they have an actual biological origin [Chai et al., 2012;Keller et al., 2013;Liang et al., 2012]. Furthermore, negative synchronizations have been observed in Parkinson's disease where it has been hypothesized that they might represent a compensatory mechanism [Peraza et al., 2017]. Figure S2: Mean z scores for all clusters showing a significant difference between DLB and HC and DLB and AD (see Figure 2 and Table II). In each boxplot the central line corresponds to the sample median, the upper and lower border of the box represent the 25 th and 75 th percentile, respectively, and the length of the whiskers is 1.5 times the interquartile range, outliers are shown by +. In addition to investigating correlations with mean connectivity within a cluster, we also tested voxelwise correlations with clinical scores. To this end, the dual regression z-scores for all DLB participants were concatenated in one 4D image and correlations with clinical scores were tested using a GLM in FSL with the respective clinical score as covariate in the design matrix. Statistical significance was assessed using randomize with 5000 permutations including a mask for the significant clusters from the HC-DLB and AD-DLB group comparisons. There was one cluster of 4 voxels in the right occipital fusiform gyrus belonging to the temporal network (TN-1) that showed negative correlation with the CAF total score. Additionally, there was a very small cluster of one voxel in the right precentral gyrus belonging to the right motor network (RMN-1) where connectivity was positively correlated with the CAF total score. However, none of these clusters survived FDRcorrection for multiple comparisons. We did not find any significant clusters for any of the other clinical scores.

Voxel-based morphometry analysis
To study changes in grey matter between the three groups a voxel-based morphometry analysis was conducted using DARTEL in SPM12 using age, gender, and total intracranial volume as covariates.
The AD group showed clusters of reduced grey matter compared to controls, mainly in right and left hippocampus (Supplementary Figure S3). No regions showed increased grey matter in AD compared to controls.
The DLB group had reduced grey matter in a small cluster in right cingulate (Supplementary Figure   S4). Again, there were no areas of increased grey matter in DLB compared to controls. There was also no difference in grey matter between the two dementia groups. Figure S3: Results from VBM analysis comparing AD and controls. Figure S4: Results from VBM analysis comparing DLB and controls.

Effect of dichotomous study covariate
To study the effect of the inclusion of the dichotomous study covariate, we repeated the dual regression analysis without the covariate and compared the results. Group differences were found in the same networks (see Supplementary Figure S5), the only difference being that some clusters became a bit larger (e.g. basal ganglia network) or smaller (e.g. insular networks) when including the covariate. Nevertheless, we decided to include the covariate in our main analysis to account for small differences between the two studies in terms of recruitment and imaging protocol. Figure S5: Dual regression results for comparison between DLB and HC without the addition of a dichotomous covariate for study membership. RSN maps are shown in red-yellow and significant group differences are shown in blue/green. All images shown in radiological convention.