Intraoperative neural signals predict rapid antidepressant effects of deep brain stimulation

Abstract

Deep brain stimulation (DBS) of the subcallosal cingulate (SCC) is a promising intervention for treatment-resistant depression (TRD). Despite the failure of a clinical trial, multiple case series have described encouraging results, especially with the introduction of improved surgical protocols. Recent evidence further suggests that tractography targeting and intraoperative exposure to stimulation enhances early antidepressant effects that further evolve with ongoing chronic DBS. Accelerating treatment gains is critical to the care of this at-risk population, and identification of intraoperative electrophysiological biomarkers of early antidepressant effects will help guide future treatment protocols. Eight patients underwent intraoperative electrophysiological recording when bilateral DBS leads were implanted in the SCC using a connectomic approach at the site previously shown to optimize 6-month treatment outcomes. A machine learning classification method was used to discriminate between intracranial local field potentials (LFPs) recorded at baseline (stimulation-naïve) and after the first exposure to SCC DBS during surgical procedures. Spectral inputs (theta, 4-8 Hz; alpha, 9-12 Hz; beta, 13-30 Hz) to the model were then evaluated for importance to classifier success and tested as predictors of the antidepressant response. A decline in depression scores by 45.6% was observed after 1 week and this early antidepressant response correlated with a decrease in SCC LFP beta power, which most contributed to classifier success. Intraoperative exposure to therapeutic stimulation may result in an acute decrease in symptoms of depression following SCC DBS surgery. The correlation of symptom improvement with an intraoperative reduction in SCC beta power suggests this electrophysiological finding as a biomarker for treatment optimization.

Conflict of interest statement

HSM reports consulting and intellectual licensing fees from Abbott Labs. PR-P has served as a consultant for Janssen Pharmaceuticals. REG serves as a consultant to Medtronic, which manufactures products related to the research described in this paper and receives compensation for these services. The terms of these arrangements have been reviewed and approved by Emory University and the Icahn School of Medicine in accordance with their respective conflict of interest policies. JTG provided consulting services for AlphaOmega Engineering during the period of the project. The remaining authors declare no competing interests.

© 2021. The Author(s).

Figures

Fig. 1. Experimental design and analytic procedures.

A Schematic of bilateral leads implanted in the subcallosal cingulate. B Timeline of intraoperative procedures. Blue boxes (PRE, MID, POST) indicate epochs of SCC local field potentials (LFP) recording. C Decrease in symptom severity measured with Hamilton Depression Rating Scale (HDRS-17) following surgical implantation. D Machine learning (ML) analysis pipeline: classification models were developed from six LFP spectral features per subject to discriminate between PRE and POST epochs. ML outputs included area under the curve (AUC) and feature importance statistics.

Fig. 2. Acute changes in brain electrophysiology following exposure to intraoperative SCC-DBS.

A Logistic regression classifier with elastic-net regularization discriminated between baseline (PRE) and post-stimulation (POST) LFPs in the SCC region, AUCmean = 0.729; SD = 0.034, N = 7. B A feature importance score indicated the relative contribution of features to classifier success (PRE vs. POST).

Fig. 3. Decrease in beta power following intraoperative exposure to bilateral DBS at tractography-defined “optimal” contacts predicts early antidepressant effects.

A Change in the left hemisphere beta power before (PRE: blue) and after (POST: red) exposure to DBS, F(1,7) = 96.23, pcorrected < 0.001. B Left hemisphere LFP beta power decrease following intraoperative exposure to “optimal” DBS correlates with greater decrease in symptom severity 1 week following surgery, r(8) = 0.730, pcorrected = 0.04. HDRS Hamilton Depression Rating Scale. C Intraoperative SCC-LFP power spectral density (left hemisphere) in single subjects before exposure to DBS (PRE) and after (POST). Subject-level power spectral density (PRE: blue, POST: red). Black boxes indicate time window of beta power feature (13–30 Hz).

Fig. 4. Classification analysis to specify the stimulation protocol to enhance acute intraoperative electrophysiological change.

A Group-level logistic regression classifiers with elastic-net regularization failed to discriminate above chance between pre-stimulation baseline (PRE) and mid-procedure (MID) LFPs, AUCmean = 0.574; SD = 0.04, N = 7 mid-procedure samples (MID) followed non-specific, unilateral testing at all contacts, including tractography-defined non-optimal contacts. B Group-level logistic regression classifiers reliably discriminated MID from post-stimulation (POST) LFP samples, where POST recording followed exposure to bilateral stimulation at the tractography-defined “optimal” contacts, AUCMEAN = 0.684; SD = 0.059, N = 7. C Patient-level logistic regression classifier performance improved following exposure to tractography-defined “optimal” stimulation (p < 0.001).