Targeting intracranial electrical stimulation to network regions defined within individuals causes network-level effects

Abstract

Intracranial electrical stimulation (ES) is routinely used therapeutically, diagnostically, and to provide causal evidence in neuroscience studies. However, our understanding of the brain network-level effects of ES remains limited. We applied precision functional mapping (PFM), based on functional magnetic resonance imaging (fMRI), to define large-scale networks within individual epilepsy patients. We show that single-pulse electrical stimulation (SPES) and high-frequency electrical stimulation (HFES) are more likely to evoke within-network responses and elicit network-related behavioral effects, respectively, when applied near to a PFM-defined network region. Network-level effects were more likely when stimulating sites in white matter, in close proximity to the targeted network, and within a region predominantly occupied by the targeted network. Further, network-specific modulation may be achievable by applying lower current intensities at these sites. Our findings support that modulation of specific networks is achievable by targeting ES to a functional anatomic "sweet spot" that can be identified using PFM.

Fig. 1:. Study overview and derived metrics.

(A) Timeline of study. Epilepsy surgery patients were recruited for a precision functional mapping (PFM) study where large-scale networks were defined in each individual prior to intracranial electrical stimulation (ES). Each participant underwent up to four magnetic resonance imaging (MRI) sessions including functional MRI resting-state runs for network definition using functional connectivity (FC). Seed- and parcellation-based maps are shown for example patient P1. Subdural (electrocorticography or ECOG) and depth (stereo electroencephalography or sEEG) electrode locations were chosen strictly for clinical purposes and were determined using a post-operative computed tomography (CT) scan. Intracranial ES was applied at different bipolar stimulation sites, current intensities (1–15mA), and frequencies (SPES: 0.5/1 Hz; HFES: 50 Hz). We conducted a post-hoc analysis of which factors influenced network-level effects of intracranial ES, focusing on (B) the Euclidean distance between the stimulation site (bipolar centroid) and the nearest gray-white matter boundary (defined as negative if in white matter), (C) the shortest (Euclidean) distance from the stimulation site to each PFM-defined network region; and (D) the “Dominance” of each network at the stimulation site, defined by centering a 10-mm full-width at half-maximum (FWHM) Gaussian on the bipolar stimulation site, then calculating the percentage of weighted vertices belonging to each network. We assessed how these stimulation site properties affected network-level evoked responses and behavioral effects of intracranial ES.

Fig 2:. Precision functional mapping estimates of large-scale networks and electrode coverage in 6 example subjects (P1-P6).

The remaining 5 subjects, which included right hemisphere and bilateral implants, are shown in Supp. Figs. S6 & S4. Precision functional mapping was used to define large-scale networks in each individual and allowed the localization of electrodes within specific networks. Inflated representations of the left cortical hemisphere are shown in lateral (left) and medial (right) views for a subset of participants (P1 – P6) (Top) Individualized network estimates were derived using data-driven clustering data (65) of the extensively collected fMRI data. The initial network maps (shown in Supp. Fig. S2) were then processed post-hoc to remove speckling (small clusters indicating noise) and to combine two networks, somatomotor network A (SMOT-A) and -B (SMOT-B), to homogenize these across individuals. (Bottom) Locations of implanted electrodes are represented using a 10-mm FWHM Gaussian centered on each bipolar contact pair. *Patient P1 had two surgeries which are combined here. DN-A: default network A, DN-B: default network B, LANG: language network, AUD: auditory network, SAL: salience network, FPN: frontoparietal control network, CON-A: cingulo-opercular network A, CON-B: cingulo-opercular network B, SMOT-A: somatomotor network A, SMOT-B: somatomotor network B, PM-PPr: premotor-posterior parietal rostral network, dATN-A: dorsal attention network A, dATN-B: dorsal attention network B, VIS: visual network.

Fig. 3:. The number of networks activated by SPES depends on the applied current intensity and the depth of the stimulation site into white matter.

(A) Recording sites were assigned to the most dominant network in the surrounding region using a winner-takes-all approach. (B) Recording sites within 20 mm (Euclidian) of the stimulation site were excluded, and if at least one distal recording site in a given network showed a significant response (trial-averaged maximum amplitude, 20–500 ms post-stimulation, Z > 2), the network was counted as being activated. (C) Response matrices for distal recording sites (rows) following SPES applied at time 0 ms at the same stimulation site but at three separate current intensities (2, 5, 7 mA). Colored circles and thin bars on the left of each matrix denotes network membership of each recording site. Black stripes on the right denote a significant response. (D) Across stimulation sites, increasing current intensity led to more networks being activated, with a plateau after ≥4 mA. Data were fit to a generalized linear model (red curve) and performance was compared to a constant model (top). A Tukey-Kramer test showed no significant difference (p > 0.05) among higher current intensities, except between 5–6 mA. E) SPES response matrices for 5 mA stimulation at three stimulation sites of varying depths into white matter (mm). (F) Across instances of 5mA stimulation, stimulation sites further into/nearer white matter caused more networks to be activated. Results of a Pearson’s correlation are shown in the top right corner.

Fig. 4:. Single-pulse electrical stimulation (SPES) led to distant activation of networks connected to the stimulation site.

The Euclidean distance from the stimulation site to a targeted network region was calculated, and the evoked activity pattern was categorized as either showing responses only within the targeted network (A, “Network N Only”), within multiple networks including the targeted network (B, “Includes Network N”), within one or more networks excluding the targeted network (C, “Other Network(s)”), or not evoking any responses (“None”). This was repeated multiple times per stimulation site for each network in which there were distant recording sites. (D) Count (left column) and proportion (right) of evoked responses in each category binned by distance from the stimulation site to the targeted network. Each plot shows a different applied current intensity. The likelihood of activating a targeted network (red and bright red colors; i.e., Network N Only, Includes Network N) decreased with increasing distance of the stimulation site to the targeted network. A permutation test showed that only stimulation sites within 10 mm of the targeted network were significantly more likely to cause targeted network effects (i.e., Network N Only and Includes Network N) than expected by chance (see asterisks on right column, p < 0.05). Current intensity-distance bins with fewer than 5 observations were not assessed (grayed out). Few (16/87) stimulation sites caused network-specific activation (bright red; i.e., Network N Only), and these were within short distances (<15 mm) of the targeted network, and more prevalent at low current intensities (~1 mA).

Fig. 5:. Network-specific responses evoked by low current intensity (1mA) single-pulse electrical stimulation (SPES) are more likely when stimulation is applied near brain regions dominated by the targeted network.

Exploratory findings focused on the few instances of low current intensity (1mA) SPES in our dataset revealed that (A) depth into white matter, (B) distance to the targeted network, and (C) dominance of the targeted network at the site of stimulation may be important factors in determining the likelihood of evoking distant activation of only the targeted network (Network N Only). Stimulation sites causing targeted network-specific responses (Network N Only) tended to be in white matter, near the targeted network, and in brain regions dominated by the targeted network. However, statistics were inconclusive, likely due to the few (26) sites where low current stimulation was applied in our dataset. Pairwise comparisons using a Tukey-Kramer test are shown at the top of each plot (*p < 0.05, ***p < 0.001). The lower panels show example stimulation sites where 1 mA SPES was applied in white matter near regions where (D) one network showed high Network Dominance, leading to network-specific activation of the dominant network, and (E) no network was clearly dominant, leading to nonspecific effects including activation of the targeted network.

Fig. 6:. Behavioral effects of high-frequency electrical stimulation (HFES) are influenced by the depth of the stimulation site into white matter.

Behavioral effects were categorized as either having elicited behavioral effects (“Behavioral Effects”), or not eliciting effects even at the maximum current intensity that could be applied (“No Effects”, max for depth electrodes: 7mA, max for grid/strip electrodes: 15 mA). Sites that led to seizures, auras, or with behavioral effects concurrent with after discharges were excluded. (A) Examples of HFES performed in participant P2 with language testing near a PFM-defined language network (LANG) region in the lateral temporal cortex (yellow). The site that was closest to gray matter (depth of −0.36 mm) did not elicit effects at 7 mA, while the other sites elicited effects at 3 mA. (B) Across all tested sites, HFES sites that elicited behavioral effects were further into/nearer white matter than those that did not (one-sided t-test; **p < 0.01). (C) Separating data points by applied current intensity, HFES sites eliciting effects at lower current intensities (~1mA) appeared to be confined to white matter within 4 mm of the boundary with gray matter, whereas HFES sites not eliciting effects until higher current intensities appeared to be more spread out, though this difference was not statistically significant. Note the higher incidence of gray matter stimulation which elicited no effects despite using higher current intensity.

Fig. 7:. High-frequency electrical stimulation (HFES) elicits behavioral effects more often when applied near behavior-relevant networks.

Focusing on HFES sites within 5 mm of gray matter (≥ −5 mm depth into white matter) which are near the functionally mapped networks, HFES effects were assessed in terms of their Euclidean distance to six large-scale networks. (A) Example participants showing the location of sites showing clear network-related effects for each network. Euclidean distance to each stimulation site is shown in the red-yellow colormap, with the relevant network boundaries (SMOT, LANG, etc.) in black. Note that these distance maps can appear discontiguous due to the surface projection step, particularly for contacts that were deeper into the brain. (B) Plot showing the count (left) and proportion (right) of different behavioral effect categories at each distance to the targeted network. Stimulation sites that were closer to the targeted network showed higher incidence network-related behavioral effects (“Clear/Mixed Network N Effect”). Only stimulation sites within 5 mm of the targeted network were significantly more likely to cause targeted network effects than expected by chance (see asterisks on right column, derived from a permutation test, p < 0.05).

Fig. 8:. Stimulation sites where high-frequency electrical stimulation (HFES) elicited single-domain behavioral effects were more dominated by the targeted network.

Stimulation sites that elicited behavioral effects related to the targeted network (Clear Network N Effect, Mixed Network N Effect) tended to be located (A) in white matter near the gray matter, and (B) close to the targeted network. Furthermore, (C) stimulation sites that elicited single-domain behavioral effects related to the targeted network (Clear Network N Effect) tended to be dominated by the targeted network. This indicates that the functional organization of the stimulated site influences what behavioral effects are elicited. Pairwise comparisons using a Tukey-Kramer test are shown at the top of each plot (*p < 0.05, **p < 0.01, ***p < 0.001).

Fig. 9:. Stimulation sites where high-frequency electrical stimulation (HFES) elicited network-related behavioral effects were closer to the relevant individualized network than a group-level network analog.

(A) For each behavioral effect type, group-level network analogs were chosen for each individualized network based on spatial overlap. Group-level network analogs are labeled by their name and color from (21). Boundaries of the individualized network are in black. (B) Stimulation sites where HFES elicited network-related behavioral effects were then assessed based on whether they were closer to the relevant individualized network (negative difference in distance) or the group-level network analog (positive difference in distance). The maps from both group and individualized approaches overlapped at most sites (difference in distance equal to 0 mm). However, sites that elicited network-related behavioral effects overall tended to be closer to the individualized networks (i.e., the distribution was shifted towards negative values).

Fig. 10:. Proposed framework for network-targeted intracranial electrical stimulation.

(A) Apply stimulation using low current intensities to reduce the volume of stimulated tissue and increase the likelihood of modulating a single network. (B) Stimulation should be applied within white matter but immediately near (e.g., < 5 mm away) to the gray matter ribbon (i.e., the “sweet spot”) to increase the likelihood of eliciting effects, possibly due to activation of axons leaving the nearby gray matter region. (C) Apply stimulation near a region of the targeted network to increase the likelihood of activating that network and/or eliciting behavioral effects associated with that network. (D) Apply stimulation in a region that is more dominated by the targeted network to increase the likelihood of achieving network-specific activation and effects.