The modulatory effect of adaptive deep brain stimulation on beta bursts in Parkinson's disease

Abstract

Adaptive deep brain stimulation uses feedback about the state of neural circuits to control stimulation rather than delivering fixed stimulation all the time, as currently performed. In patients with Parkinson's disease, elevations in beta activity (13-35 Hz) in the subthalamic nucleus have been demonstrated to correlate with clinical impairment and have provided the basis for feedback control in trials of adaptive deep brain stimulation. These pilot studies have suggested that adaptive deep brain stimulation may potentially be more effective, efficient and selective than conventional deep brain stimulation, implying mechanistic differences between the two approaches. Here we test the hypothesis that such differences arise through differential effects on the temporal dynamics of beta activity. The latter is not constantly increased in Parkinson's disease, but comes in bursts of different durations and amplitudes. We demonstrate that the amplitude of beta activity in the subthalamic nucleus increases in proportion to burst duration, consistent with progressively increasing synchronization. Effective adaptive deep brain stimulation truncated long beta bursts shifting the distribution of burst duration away from long duration with large amplitude towards short duration, lower amplitude bursts. Critically, bursts with shorter duration are negatively and bursts with longer duration positively correlated with the motor impairment off stimulation. Conventional deep brain stimulation did not change the distribution of burst durations. Although both adaptive and conventional deep brain stimulation suppressed mean beta activity amplitude compared to the unstimulated state, this was achieved by a selective effect on burst duration during adaptive deep brain stimulation, whereas conventional deep brain stimulation globally suppressed beta activity. We posit that the relatively selective effect of adaptive deep brain stimulation provides a rationale for why this approach could be more efficacious than conventional continuous deep brain stimulation in the treatment of Parkinson's disease, and helps inform how adaptive deep brain stimulation might best be delivered.

Keywords: Parkinson’s disease; basal ganglia; beta oscillations; closed-loop control; deep brain stimulation.

Figure 1

Steps of burst determination. (A) The analogue LFP signal was filtered around individual beta peak frequency (Table 1). The signal was rectified and smoothed to obtain the envelope of the beta activity. For each condition a threshold was then set at the 75th percentile of the beta amplitude. The onset of a burst was defined as when the rectified signal crossed the threshold amplitude and the end of the burst defined as when the amplitude fell below threshold. (B) All bursts with a duration longer than 100 ms were considered. Bursts were further categorized according to their duration into nine time windows (see ‘Materials and methods’ section). Example of burst scatterplot with burst shown up to 1000 ms (Subject 3, left side).

Figure 2

Beta burst distributions during noStim, adaptive and conventional DBS. (A) Number of bursts of different durations during noStim, adaptive and conventional DBS, where bursts are defined as periods of beta activity that exceed the 75th percentile for longer than 100 ms. For adaptive DBS the number of shorter bursts (200–400 ms) is higher and the number of longer bursts (600–700 and 800–900 ms) is lower compared to noStim. The comparison between adaptive and conventional DBS similarly shows a higher number of short bursts (300–400 ms) and a lower number of long bursts (800–900 ms) during adaptive DBS. The number of bursts of different durations did not differ between noStim and conventional DBS. Values are represented as mean + SEM; ^*P < 0.05. (B) Summarizes the data from different thresholds by showing how often the number of bursts of a given time window during adaptive DBS significantly differed from those of the same time window during conventional DBS or noStim (Supplementary Fig. 1). aDBS = adaptive DBS; cDBS = conventional DBS.

Figure 3

Burst trimming effects of adaptive DBS illustrated on a series of beta bursts. Subject 11, left side. The top row shows a consecutive sequence of beta bursts observed during adaptive DBS. The bottom row shows the stimulation pattern as induced by the corresponding bursts above. Bursts appear terminated by the triggered stimulation and bursts with black arrows do end relatively abruptly, coinciding with the ramp-up of stimulation. aDBS = adaptive DBS.

Figure 4

Dichotomized distribution for short and long bursts during noStim, adaptive and conventional DBS. (A) The number of normalized bursts (categorized into short bursts: 100–600 ms and long bursts: >600 ms) for the 75% threshold, which confirms a similar pattern with a higher number of short bursts in adaptive DBS (compared to noStim) and a lower number of long bursts in adaptive DBS (compared to conventional DBS and noStim). (B) The total burst duration of the normalized bursts for the 75% threshold. This confirms a higher total time of short bursts for adaptive DBS (compared to noStim and conventional DBS) and a lower total time of long bursts in adaptive DBS (compared to noStim and conventional DBS). (C) The burst distribution as percentage normalized bursts for noStim, adaptive and conventional DBS across the various thresholds (see also Supplementary Fig. 1). The corresponding repeated measures ANOVA again confirmed an interaction between stimulation condition and burst duration [F(2,30) = 6.627, P = 0.004], the corresponding post hoc tests indicate a higher amount of shorter bursts in adaptive DBS compared to noStim [t(15) = 3.056, P = 0.024] and conventional DBS [t(15) = 2.643, P = 0.055], while the number of longer bursts is lower in adaptive DBS compared to noStim [t(15) = − 3.056, P = 0.024] and conventional DBS [t(15) = − 2.643, P = 0.055]. No difference in the distribution for short and long bursts was found between noStim and conventional DBS [t(15) = 0.188, P = 1; t(15) = − 0.188,P = 1]. (D) The percentage distribution of the total burst duration for the normalized bursts across the various thresholds. The repeated measures ANOVA confirmed again a significant interaction between condition and duration of bursts [F(1.4,21.6) = 9.648, P = 0.002]. The post hoc tests showed a longer total duration for short bursts in adaptive DBS compared to noStim [t(15) = 3.303, P = 0.014] and conventional DBS [t(15) = 3.336, P = 0.014], while the total duration for long bursts was reduced in adaptive DBS compared to noStim [t(15) = − 3.303, P = 0.014] and conventional DBS [t(15) = − 3.336, P = 0.014]. Again no difference in the total duration for short and long bursts was found between noStim and conventional DBS [t(15) = 0.641, P = 1; t(15) = − 0.641, P = 1]. Values are represented as mean + SEM; ^*P < 0.05. Asterisks in brackets: P-value significant before correction for multiple comparisons (Bonferroni) only.

Figure 5

Relationship between burst duration and burst amplitude for the representative 75% threshold. (A) The mean amplitudes for different durations for noStim, adaptive and conventional DBS. SEMs are shown for noStim only. The strong positive correlation indicates the longer the burst duration, the higher the burst amplitude. A slight flattening of this relation can be seen during adaptive DBS at longer burst durations. A second order polynomial was fitted to the data of the three conditions (see equations). Within subjects, a significant correlation could be found for almost all the hemispheres and conditions (noStim 14/16, adaptive DBS 11/16, conventional DBS 13/16). The results of the correlation analyses between burst duration and burst amplitude across all the thresholds are shown in Supplementary Fig. 5. (B) Mean burst amplitude for noStim, adaptive and conventional DBS averaged across hemispheres and time windows. Both adaptive DBS and conventional DBS show a significant reduction in beta amplitude compared to noStim. However, no difference was found between adaptive and conventional DBS. (C) Integrated burst amplitude (normalized to 100%, which corresponds to total integrated amplitude summed across all time windows) for short bursts (100–600 ms) and long bursts (>600 ms). Stimulation conditions show significantly different amplitude effects when burst duration is considered. Adaptive DBS has a higher integrated amplitude in shorter bursts, while conventional DBS and noStim have a higher integrated amplitude in longer bursts. Supplementary Fig. 7 illustrates mean amplitude as well as the integrated amplitude across the different thresholds. Values are represented as mean + SEM; ^*P < 0.05, ^**P < 0.01, ^***P < 0.001. aDBS = adaptive DBS; cDBS = conventional DBS.

Figure 6

Clinical correlation between burst duration and clinical impairment. (A) Pearson’s correlations between clinical impairment and the percentage amount of bursts during different burst time windows across the various thresholds. These show that shorter bursts tend to be negatively correlated with clinical impairment and longer bursts tend to be positively correlated with clinical impairment. (B) Example scatter plot of percentage amount of short bursts (of 200–300 ms duration) and clinical impairment (UPDRS Part III items 20, 22 and 23 contralateral to the recording side). (C) Example scatter plot of percentage amount of long bursts (700–800 ms) and clinical impairment. B and C are data for representative threshold (75th percentile).

Figure 7

Summary schematic. Beta amplitude profile shown during noStim, adaptive and conventional DBS. Without DBS both short and long beta bursts occur. During adaptive DBS the longer bursts are trimmed. This in turn affords more space for shorter bursts to occur. During conventional DBS the distribution of short and long bursts does not change, but overall beta is still suppressed, implying that the amplitude of all beta bursts is reduced. Note that background levels of beta are shown as similar between conditions as our signal processing is focussed on burst behaviour. We cannot rule out additional changes in background levels of beta in the subthalamic nucleus, particularly during conventional DBS, but such changes are difficult to ascertain whilst amplifier noise floors may vary between conditions. aDBS = adaptive DBS; cDBS = conventional DBS.