Mapping dysfunctional circuits in the frontal cortex using deep brain stimulation

Abstract

Frontal circuits play a critical role in motor, cognitive and affective processing, and their dysfunction may result in a variety of brain disorders. However, exactly which frontal domains mediate which (dys)functions remains largely elusive. We studied 534 deep brain stimulation electrodes implanted to treat four different brain disorders. By analyzing which connections were modulated for optimal therapeutic response across these disorders, we segregated the frontal cortex into circuits that had become dysfunctional in each of them. Dysfunctional circuits were topographically arranged from occipital to frontal, ranging from interconnections with sensorimotor cortices in dystonia, the primary motor cortex in Tourette's syndrome, the supplementary motor area in Parkinson's disease, to ventromedial prefrontal and anterior cingulate cortices in obsessive-compulsive disorder. Our findings highlight the integration of deep brain stimulation with brain connectomics as a powerful tool to explore couplings between brain structure and functional impairments in the human brain.

Fig. 1. Overview of the twofold group-level approach of (sub)cortical dysfunction mapping.

a, DBS Sweet Spot Mapping. Patient-specific electrode reconstructions were first derived relative to their precise position within the STN region and integrated with individual stimulation parameters to estimate E-field magnitudes. Subsequently, Spearman’s rank correlations between E-field magnitudes and clinical improvements were performed (separately for each disease). Applying this procedure across voxels resulted in a detailed grid of positively (sweet spot) and negatively (sour spot, not shown here) associated stimulation sites. b, DBS Fiber Filtering. Each streamline within a predefined normative connectome was weighted by its ability to discern good from poor responders in each respective cohort. To do so, the peak E-field magnitudes among samples drawn along the course of each streamline were Spearman’s rank correlated with clinical outcomes. Streamlines predominantly modulated by high E-field magnitudes of good responders received high positive weights (sweet streamlines), whereas those associated with high E-field magnitudes of poor responders were attributed high negative weights (sour streamlines, not represented here).

Fig. 2. Overview of electrode placements relative to the STN across discovery cohorts.

Left panels, DBS electrode placement is shown in relation to a posterior view of the STN in DYT (n = 56), PD (n = 94), TS (n = 14) and OCD (n = 19) cohorts, respectively. Electrode contacts are visualized as point clouds. Right panel, visualization of DBS leads of the four discovery cohorts investigated in the present study are featured in the axial plane and colored according to indication. The STN is defined by the DISTAL atlas, version 1.1 (ref. ), with an axial plane of the BigBrain template in 100-µm resolution displayed as a backdrop (y = −5 mm, z = −10 mm).

Fig. 3. Segregation of dysfunction mappings at the subthalamic level by disease-specific stimulation effects.

Middle panel (center), the topographical organization of disorder-specific DBS sweet spots in DYT (n = 56), TS (n = 14), PD (n = 94) and OCD (n = 19) is shown as a density cloud plot relative to a three-dimensional model of the left STN in template space derived from the DISTAL atlas, version 1.1 (ref. ). Sphere size and transparency indicate Spearman’s rank correlation strength between stimulation impact and clinical improvements at a given coordinate, with bigger and less transparent spheres coding for higher correlations. Below, binarized and thresholded sweet spot peaks are projected onto the STN surface. Top and bottom panels, axial and coronal views of sweet and sour spots are displayed relative to the left STN (black outlines), independently for each disorder, superimposed onto a 100-µm ex vivo brain template. Voxels are color-coded by degree of Spearman’s rank correlation (warm colors for positive associations and cool colors for negative associations) between E-field magnitudes and clinical improvements. Middle panel (left and right), Spearman’s correlation plots show amounts of clinical outcome variance explained by spatial similarity of E-field peaks with disease-wise sweet spot models (expressed as Sweet Spot Score under each E-field, averaged for bilateral scores per patient) across the cohort (two-sided tests). Gray shaded areas are representative of 95% confidence intervals.

Fig. 4. Disease-specific sweet streamline models in each discovery cohort.

a, Sweet streamlines in DYT (n = 56; peak R = 0.36), PD (n = 94; peak R = 0.37), TS (n = 14; peak R = 0.73) and OCD (n = 19; peak R = 0.49) associated with beneficial stimulation outcomes were filtered from a population-based group connectome. The top row demonstrates the set of connections (in white) seeding from stimulation volumes across patients in each of the four disorders. Among these plain connections, only those were isolated via DBS Fiber Filtering (middle row) whose modulation was Spearman’s rank correlated with clinical outcomes (bottom row). Sweet streamlines are highlighted in disease-specific color and displayed in thresholded and binarized fashion. Results are shown against a sagittal slice (x = −5 mm) of the 7T MRI ex vivo 100-µm human brain template, in conjunction with a three-dimensional model of the right STN in template space from the DISTAL atlas, version 1.1 (ref. ). b, In-sample correlations and fivefold CVs are reported for models informed on four different normative connectomes. Plots in the top row represent the fitting of a linear model to determine the degree to which the overlap of E-field magnitudes with selected HCP 985 Connectome sweet streamlines explains variance in empirical clinical outcome across the cohort, as calculated using Spearman’s correlation (two-sided tests). The magnitude of E-field overlap with sweet streamline models in this analysis is expressed as weighted peak 5% of Fiber R scores under each E-field, averaged across bilateral scores per patient. Gray shaded areas indicate 95% confidence intervals.

Fig. 5. Topography of streamlines and interconnected cortical sites associated with therapeutic stimulation effects.

a, Segregation into therapeutic networks is achieved by means of DBS Fiber Filtering in DYT (n = 56), PD (n = 94), TS (n = 14) and OCD (n = 19). Disease-specific optimal streamlines were isolated from a high-resolution normative group connectome through association with clinical effects in each disorder. This was achieved using Spearman’s rank correlation between peak E-field magnitudes by which each streamline was modulated and clinical improvements across the cohort. Mappings are displayed in thresholded form, against a sagittal slice (x = −5 mm) of a brain cytoarchitecture atlas in ICBM 2009b Nonlinear Asymmetric (‘MNI’) space. Streamline color intensity is representative of R value magnitude, with darker colors corresponding to higher correlations. b, The same streamlines are shown in conjunction with a transparent brain in template space along with delineations that are color-coded by disease. c, To derive the cortical topography of dysfunction mappings, smoothed, thresholded and binarized density maps of sweet streamlines were projected onto a brain template in MNI space. Circles show close-up views of disease-wise interconnected cortical sites, anatomically characterized based on the JHU atlas parcellation. Legend of relevant regions, with corresponding JHU atlas denominators in brackets: 1 (JHU: 23 and 24), postcentral gyrus; 2 (1 and 2), superior frontal gyrus (posterior segment); 3 (3 and 4), superior frontal gyrus (prefrontal cortex); 4 (25 and 26), precentral gyrus; 5 (5 and 6), superior frontal gyrus (frontal pole); 6 (9 and 10), middle frontal gyrus (dorsal prefrontal cortex); 7 (17 and 18), lateral fronto-orbital gyrus; 8 (13 and 14), inferior frontal gyrus pars orbitalis; 9 (15 and 16), inferior frontal gyrus pars triangularis.

Fig. 6. Retrospective and prospective validations of therapeutic streamline targets.

To probe the validity of PD and OCD streamline models, five validation experiments were carried out. a, First and second, empirical outcomes of two additional independent datasets (PD: n = 32 and OCD: n = 35) could significantly be estimated based on the degree of overlap of their stimulation volumes with the streamline models. Sweet streamline models calculated on discovery cohorts are represented in disease-specific color, in thresholded and binarized fashion. Model validity is expressed in the form of Spearman’s correlations between the stimulation magnitude by which positive streamlines in the model were respectively modulated (weighted peak 5% of Fiber R scores attributed to each E-field, averaged for bilateral scores per patient) and empirical clinical improvements across the cohort (two-sided tests). Gray shaded areas represent 95% confidence intervals. b, Third and fourth, prospective reprogramming was undertaken in two patients. In the patient with PD, directional electrodes had been implanted, so the current was divided using a 70/30% rule based on the contacts with the strongest and second-to-strongest streamline overlaps. This led to an improvement of 71% on the UPDRS-III compared to 60% using clinical settings. In the OCD case, the contact was selected based on visual inspection with the streamline model by the clinical team. This led to a reduction of 37% on the Y-BOCS compared to 17% under clinician-selected parameters. c, Fifth, a prospective OCD case underwent streamline-guided DBS surgery. Electrodes were activated at the contact with the highest streamline overlaps (most ventral contacts bilaterally), leading to a rapid Y-BOCS reduction of 77% already 1 month after surgery. Depending on the respective target, reconstructed electrodes and stimulation volumes are featured relative to three-dimensional models of the STN from the DISTAL atlas, version 1.1 (ref. ), or of the nucleus accumbens (Nac) from the California Institute of Technology reinforcement learning (CIT168) atlas, version 1.1 (ref. ), and against anatomical slices of a 100-µm ex vivo brain template. PREOP, pre-operative; PT., points.

Fig. 7. Conserved segregation of dysfunction mappings among indirect pallido-subthalamic connections.

Disease-wise sweet streamlines retain a high degree of specificity along their indirect pathway trajectory connecting the STN with the GPi and GPe. Connectivity is modeled based on the Basal Ganglia Pathway Atlas. Sweet streamlines associated with optimal DBS outcomes in DYT (n = 56) are interconnected with sensorimotor (a), in TS (n = 14) with associative (b), in PD (n = 94) with premotor (c) and in OCD (n = 19) with limbic (d) STN territories. Streamlines are thresholded and represented in disease-specific color. Color intensities attributed to each streamline code for the degree of Spearman’s rank correlation between streamline modulation (peak E-field magnitudes) and clinical outcomes across the disease cohort, with darker colors indicative of higher correlations. Results are displayed relative to several anatomical structures from the DISTAL atlas, version 1.1 (ref. ), and in conjunction with an axial slice (z = −10 mm) of the BigBrain template. ASSOC. STN, associative territory of the STN; LIMB. STN, limbic territory of the STN; MOTOR STN, motor territory of the STN; RN, red nucleus.