A Longitudinal Magnetoencephalographic Study of the Effects of Deep Brain Stimulation on Neuronal Dynamics in Severe Anorexia Nervosa

Abstract

Anorexia Nervosa (AN) is a debilitating psychiatric disorder characterized by the relentless pursuit of thinness, leading to severe emaciation. Magnetoencephalography (MEG)was used to record the neuronal response in seven patients with treatment-resistant AN while completing a disorder-relevant food wanting task. The patients underwent a 15-month protocol, where MEG scans were conducted pre-operatively, post-operatively prior to deep brain stimulation (DBS) switch on, twice during a blind on/off month and at protocol end. Electrodes were implanted bilaterally into the nucleus accumbens with stimulation at the anterior limb of the internal capsule using rechargeable implantable pulse generators. Three patients met criteria as responders at 12 months of stimulation, showing reductions of eating disorder psychopathology of over 35%. An increase in alpha power, as well as evoked power at latencies typically associated with visual processing, working memory, and contextual integration was observed in ON compared to OFF sessions across all seven patients. Moreover, an increase in evoked power at P600-like latencies as well as an increase in γ-band phase-locking over anterior-to-posterior regions were observed for high- compared to low-calorie food image only in ON sessions. These findings indicate that DBS modulates neuronal process in regions far outside the stimulation target site and at latencies possibly reflecting task specific processing, thereby providing further evidence that deep brain stimulation can play a role in the treatment of otherwise intractable psychiatric disorders.

Keywords: N400 & P600; alpha power; anorexia nervosa; deep brain stimulation; magnetoencephalography; phase-locking; treatment.

Figure 1

Taskand analytical approach. (A) The graph illustrates one experimental trail comprising image display, cue display, andinter-stimulus interval. Sample images of the low- and high-calorie categories are shown on the right. (B)Analysis pipeline. MaxFilter^TM is a proprietary software of MEGIN (TSSS temporal extension of signal-space-separation). ¹High amplitudes detection based on global field power. Bad intervals were zeroed (resting data) or removed from the epoch-based analysis (total loss < 5%). ²Event related field power was used as a measure of activity. The signals were low-pass (<=30 Hz) filtered, baseline corrected (−100–0 ms), and squared before averaging over trials. N_e: number of epochs. ³Phase-locking (PL) based on a Gabor transform with resolutions Δt_{80 Hz} ≈11 ms and Δf_{80 Hz} ≈7 Hz in time and frequency, respectively. The i^th-epoch Gabor spectral coefficient is denoted by c_i. The variance of S was estimated using a bootstrap (over epochs) algorithm with 250 repetitions. Note PL is unit-less. The features (measures) are defined for each subject, experimental condition (t, f, or t-f) point, and channel. ⁴The channel statistics yield probabilities τ defined for each (t, f, or t-f) point and channel, given a feature and comparison of interest. ⁵The global (whole head; N_c: number of channels) statistics are defined for each (t, f, or t-f) point. Intervals with p <= 0.01 were considered significant and mapped back to the channel level (integrating over time and/or frequency if appropriate) for further consideration.

Figure 2

Effect of deep brain stimulation on resting α-activity. (A) The panel shows the spatial distribution of significance of the difference in (grand-mean) α-power (at 9 Hz) between conditions, where activity is larger in ON compared to OFF sessions (middle panel). The relative increase in activity over anterior and posterior regions is about 0.3 dB and 0.5 dB, respectively. (B) The panel shows OFF and ON spectra obtained in one patient. Note these are raw spectra in which line noise has not been attenuated. The inset shows a 2D projection of the MEG channels (right ear on right, front at top).

Figure 3

Effectsof deep brain stimulation and calorific value on evoked responses. (A) Shown is the global evoked power summed over all subjects, trials, and channels. The gray bars indicate time intervals where a significant modulation of the neuronal response was observed. The inset shows differential (ON–OFF) activity maps corresponding to the effect at 190 ms (highlighted in dark gray in the graph) for non-responders and responders. The circles indicate a putative lateralization of activity in responders during stimulation. (B) The maps show the distribution of significance (upper row) corresponding to differential effects of stimulations at three latencies and one interaction effect between stimulation and image type. The (differential) activity maps shown in the lower row are based on evoked power (squared ERF) calculations, however, for visual presentation, the (grand-mean) data have been transformed back to the physical unit (fT/cm) of the MEG gradiometers. Note that some regions of significance exhibit only small differential effects and larger differences in evoked activity are not significant. The insets (bottom row) illustrate the consistency of results when considering subsets of sessions. 100 ms: Session-3 (ON) minus Session-4 (OFF). 440 ms upper: ON minus OFF (sessions 1, 2, 5); lower: ON minus OFF (sessions 3, 4). See text for details.

Figure 4

Time-frequency planes. Phase-locking in the high γ-band was modulated by a Calorie × DBS interaction (left), as well DBS (bottom right) and Calorie effects (top right). The insets show the distributions of significance over head for each cluster, where all significant regions within a cluster corresponds to a difference in phase-locking as indicated (at grand-mean level; phase-locking values not shown). The phase-locking indicates time-dependent neuronal networks: anterior-to-posterior, mainly parietal, and mainly occipital. The clusters locate at γ1: 260–285 ms, 75–92 Hz; γ2: 420–440 ms, 71–86 Hz; γ3: 918–945 ms, 83–100 Hz; and γ4: 850–895 ms, 64–83 Hz. Note that γ2 (grayed out) was not further considered (see text for explanation). Also note that deep brain stimulation had a strong influence on phase-locking, thereby rendering the corresponding plane uninterpretable. For completeness, the time-frequency plane of Calorie × DBS significance without overlays is shown as an inset (bottom left).