A brain network for deep brain stimulation induced cognitive decline in Parkinson's disease

Abstract

Deep brain stimulation is an effective treatment for Parkinson's disease but can be complicated by side-effects such as cognitive decline. There is often a delay before this side-effect is apparent and the mechanism is unknown, making it difficult to identify patients at risk or select appropriate deep brain stimulation settings. Here, we test whether connectivity between the stimulation site and other brain regions is associated with cognitive decline following deep brain stimulation. First, we studied a unique patient cohort with cognitive decline following subthalamic deep brain stimulation for Parkinson's disease (n = 10) where reprogramming relieved the side-effect without loss of motor benefit. Using resting state functional connectivity data from a large normative cohort (n = 1000), we computed connectivity between each stimulation site and the subiculum, an a priori brain region functionally connected to brain lesions causing memory impairment. Connectivity between deep brain stimulation sites and this same subiculum region was significantly associated with deep brain stimulation induced cognitive decline (P < 0.02). We next performed a data-driven analysis to identify connections most associated with deep brain stimulation induced cognitive decline. Deep brain stimulation sites causing cognitive decline (versus those that did not) were more connected to the anterior cingulate, caudate nucleus, hippocampus, and cognitive regions of the cerebellum (PFWE < 0.05). The spatial topography of this deep brain stimulation-based circuit for cognitive decline aligned with an a priori lesion-based circuit for memory impairment (P = 0.017). To begin translating these results into a clinical tool that might be used for deep brain stimulation programming, we generated a 'heat map' in which the intensity of each voxel reflects the connectivity to our cognitive decline circuit. We then validated this heat map using an independent dataset of Parkinson's disease patients in which cognitive performance was measured following subthalamic deep brain stimulation (n = 33). Intersection of deep brain stimulation sites with our heat map was correlated with changes in the Mattis dementia rating scale 1 year after lead implantation (r = 0.39; P = 0.028). Finally, to illustrate how this heat map might be used in clinical practice, we present a case that was flagged as 'high risk' for cognitive decline based on intersection of the patient's deep brain stimulation site with our heat map. This patient had indeed experienced cognitive decline and our heat map was used to select alternative deep brain stimulation parameters. At 14 days follow-up the patient's cognition improved without loss of motor benefit. These results lend insight into the mechanism of deep brain stimulation induced cognitive decline and suggest that connectivity-based heat maps may help identify patients at risk and who might benefit from deep brain stimulation reprogramming.

Keywords: Parkinson’s disease; cognitive decline; connectivity; deep brain stimulation.

Figure 1

Cognitive decline following STN-DBS can be improved with reprogramming, but there is no significant difference in the stimulation location. DBS reprogramming resulted in significant improvement in working memory performance in a previously published cohort of 10 Parkinson’s disease patients. (A) Reproduced from Frankemolle et al.). VTAs associated with impaired working memory (B) and the reprogrammed VTAs associated with improved working memory (C) showed no significant difference in stimulation location (D). *P < 0.05.

Figure 2

Connectivity of stimulation site to the subiculum is associated with cognitive decline. For each patient, the VTA associated with cognitive decline (A, left) and the reprogrammed VTA associated with improvement (A, right) were used as seed regions for a functional connectivity analysis, leveraging a normative connectome dataset from 1000 healthy subjects (B). We computed connectivity between each VTA and an a priori region of interest in the subiculum previously linked to lesion-induced memory impairment (C, red, reproduced from Ferguson et al.). VTAs associated with cognitive decline were significantly more connected to the subiculum than reprogrammed VTAs associated with improvement (D).

Figure 3

Connections associated with DBS-induced cognitive decline are similar to connections associated with lesion-induced memory impairment. Connections associated with DBS induced cognitive decline are shown using no threshold (A) and after FWE correction for multiple comparisons (B). Connections associated with lesion-induced memory impairment are shown using no threshold (C) and after FWE correction for multiple comparisons (D) as previously reported in Ferguson et al. Warm colours denote voxels more connected to VTAs/lesions associated with cognitive impairment while cool colours denote voxels more connected to reprogrammed VTAs/lesions not associated with cognitive impairment. The unthresholded maps (A and C) can be compared using spatial correlation and permutation analysis and are significantly similar to one another (r = 0.66; P = 0.017). The FWE corrected maps (B and D) show that similar voxels are statistically significant across both analyses.

Figure 4

Cognitive decline heat map and validation in an independent cohort. To begin translating our network results into a clinical tool, we generated a heat map where the intensity at each voxels reflects the connectivity of that voxel to our cognitive decline network (A). In the area of the STN (white outline), there is an anterior-posterior and dorsal-ventral gradient. We then tested this heat map in an independent cohort of Parkinson’s disease patients with STN DBS and MDRS at baseline and 1 year after DBS (B). Intersection between VTAs and our cognitive decline heat map was corelated with cognitive decline measured at 1 year (r = 0.39; P < 0.05).

Figure 5

Using the cognitive decline heat map for risk assessment and potential reprogramming. Patients were stratified into three risk groups based on VTA overlap with our cognitive decline heat map (A). Parkinson’s disease patients in the high-risk cohort showed significantly more cognitive decline than patients in low-risk cohort and medium risk cohort. We simulated all possible monopolar electrode choices for patients in the high-risk group (B) and choose the setting that would minimize the risk of cognitive decline (C). In 8 of 11 patients there were alternative programming settings where cognitive decline could potentially be reduced. One of these high-risk patients presented for routine clinical follow-up and was reprogrammed, moving from a high risk setting to a medium-risk setting based on the heat map (D). The MDRS improved by 7 points, while parkinsonism was unchanged (48% to 52% UPDRS III reduced to baseline).

Figure 6

DBS programming prediction for an individualized treatment An exhaustive evaluation of this patients on the heat map of cognitive decline detected a preferable electrode contact combination to reduce the risk of cognitive side-effects. We called back this patient and changed stimulation settings from right lead C2 to C3. Parkinsonism was equally controlled (48–52% UPDRS III reduced to baseline) 14 days in the medication OFF state, while the MDRS improved by 7 points to previous DBS settings (still 6 points reduced to preoperative baseline).